AU2007200435B2 - Method and system for defining sets by querying relational data using a set definition language - Google Patents

Description

P/00/011 Regulation 3.2 a) C4

AUSTRALIA

Patents Act 1990 COMPLETE SPECIFICATION STANDARD PATENT

(ORIGINAL)

Name of Applicant: Actual Inventors: Address for Service: Invention Title: deCODE Genetics ehf.

Sturlugotu 8, IS-101 Reykjavik, Iceland GUDBJARTSSON, Hakon; ARNARSON, Thorvaldur, S.; ROVENSKY, Pavol; PALMASON, Vilmurdur DAVIES COLLISON CAVE, Patent Trademark Attorneys, of 1 Nicholson Street, Melbourne, 3000, Victoria, Australia Ph: 03 9254 2777 Fax: 03 9254 2770 Attorney Code: DM Method and system for defining sets by querying relational data using a set definition language The following statement is a full description of this invention, including the best method of performing it known to us:- -1- METHOD AND SYSTEM FOR DEFINING SETS BY QUERYING RELATIONAL DATA USING A SET DEFINITION LANGUAGE RELATED APPLICATIONS This application is a continuation-in-part of U.S. Application No. 10/603,112, filed June 24, 2003, which is a continuation-in-part of U.S. Application No.

10/356,365, filed January 30, 2003. The entire teachings of the foregoing applications are incorporated herein by reference.

BACKGROUND OF THE INVENTION Field of the Invention The present invention relates to the usage pattern, commonly found in many software applications, of defining sets of objects based on object attributes. A specifically designed set definition language for defining sets, abbreviated SDL, is described as well as a software system that implements this language efficiently on top of standard relational database management systems (RDBMS).

Description of Related Art Without a doubt, the most common query language is the SQL language that is implemented in most relational database systems (RDBMS). However, although SQL is a very powerful language it is too complex for many users because of its versatile nature and the need for the user to know the underlying data schema. The aim with the invention SDL language is to define a language that is powerful enough to allow users to define sets based on multiple criteria, (ii) define a language that is easy to use, and (iii) define a language that is easier to learn than SQL which does not require the user to know the underlying data schema. The SDL system is invented in the lifescience domain, nevertheless, it has a much wider applicability. Similar attempts have been made before in that domain, such as with the health query language (HQL see www.clinical info.co.uk/miquest.htm), where instead of eliminating the need for schema knowledge, the schema was kept fixed, hence allowing for certain simplifications in the language as compared to SQL. Thus, HQL is considered simple enough such that the average medical doctor can use it for epidemiological studies.

HQL is however both a language for defining patient sets and calculating statistics on those sets, whereas SDL leaves that task to more standard on-line analytical processing (OLAP) systems.

Systems that aim at providing query capabilities for users while hiding the complexity of formal query languages such as SQL are not new. For instance, tools have been devised to facilitate the generation of SQL queries with a graphical support (GUI), such as by Shaw et.al., "Apparatus and Method for Synthesizing a Query for Accessing a Relational Database", U.S. Patent No. 4,506,326. Unlike SDL, this system is purely graphical and generates QBE syntax that is then translated into SQL.

Other systems of similar nature, such as Business-Objects see Cambot and Liautaud, "Relational Database Access System Using Semantically Dynamic Objects", U.S. Patent No. 5,555,403, have been made to hide the SQL and the database schemas from the users. Like the system mentioned previously, BO does not have its own language and is as such GUI based. However, in addition to the former system it also provides an abstraction on top of the SQL metadata, the so-called business objects and rules for building database joins, given a list of such objects in a report. Database abstraction of this nature goes further back. In the work of El- Sharkawi et.al. "Architecture and Implementation of ENLI: Example-Based Natural Language Assisted Interface" Proc. of PARBASE-90, Miami, Florida, March 6-9, 1990, pp.

4 3 0 432, the authors use an English sentence to describe the meaning of each database attribute to build an English-schema. Their English-like query language is then translated into QBE before it is mapped to SQL. Other systems of similar nature are knowledge based visual query systems that map a knowledge base onto a relational database system. A paper by K. L. Siau et.al., "Visual Knowledge Query Language as a Front-End to Relational Systems," Proceedings of the 15th Annual International Computer Software and Applications Conference, 1991, pp. 373-378, Tokyo: IEEE Computer Society Press, describes a knowledge abstraction based on an enhanced entity-relationship model (EER). The system they present is a GUI-based application that uses an EER-based knowledge schema and a visual knowledge query language (VKQL) that is mapped to SQL for query evaluation.

-3- Although the invention SDL system shares some aspects with the above systems, there are also clear distinctions. First, the SDL system is based on a new language as well as novel definitions of virtual relations and dimension attributes. A new and unique feature of a set-output-dimension enables implicit relational equijoins and a very sparse and intuitive syntax. Thus in the present invention the SDL dimensions, the virtual relations and the SDL language define the relational algebraic constraints, as compared for instance to BO where the join-rules for objects are defined in explicit metadata. Also, the BO system is purely GUI based whereas the invention SDL system is centered on the SDL language. The SDL query tools use English descriptions for dimensions, i.e. a metadata on the dimensions. The descriptions are not part of the SDL language, however, they can provide a more userfriendly, English-like feeling for the SDL expression, as well as an easier view on the dimension metadata. Note that the dimensions can also have very descriptive names as long as there is no naming conflict. This can however result in longer SDL expressions. A major difference in the present invention SDL system from all of the above systems is its focus on sets and its approach for defining the sets. In the SDL language, a set is defined without any connection to the view that is used to present the elements in the set. These set views can be textual reports or graphical in nature.

Finally, the SDL language is very true to its relational origin and can therefore be integrated with the SQL language, as shown in the present invention, thereby generating a new language that Applicants refer to here as SSDL.

Another language or protocol with similar aim as the invention SDL is the lightweight directory access protocol (LDAP), see S. Shi, E. Stokes, D. Byrne, C.

Corn, D. Bachmann, and T. Jones, "An enterprise directory solution with DB2", IBM Systems Journal, 39(2): 360-383, 2000. Common to them is the set definition pattern, however, LDAP uses very different syntax than SDL aimed at application programmers and unlike SDL, LDAP also provides access control methods to its data.

Also, their data structures implementation, i.e. their approach for storing data with arbitrary attributes in a RDBMS is very different from the one presented in this invention. Implementation with data structures that are closer to the one presented here for SDL, although for strikingly different languages such as Smalltalk, can be found in work by B. Czejdo et.al., "Integration of Database Systems and Smalltalk", in Proceedings of Symposium on Applied Computing, Kansas City, 1991. Recently there has also been a large effort in defining XML query languages and work to map them into SQL. See Florescu, D. and D. Kossman, "A Performance Evaluation of Alternative Mapping Schemes for Storing XML Data in a Relational Database", Technical Report, INRIA, France, May 1999 or F. Tian et.al. "The Design and Performance Evaluation of Alternative XML Storage Strategies", ACM Sigmod Record, vol. 3 2002 Some of the storage strategies described in the above references are similar to the one used in the present invention, however, SDL is significantly different from all these languages.

There are certain features in the SDL language, such as primary and virtual dimensions, that have some correspondence to object oriented techniques (see R.G.G.

Cattell, "Object Data Management: Object-Oriented and Extended Relational Database Systems", 1994, Addison Wesley Publishing Company, Inc.). In object schemas that utilize composition or aggregation, it is a common technique in objectoriented languages to de-reference references with path expressions (cascaded dot notation). Object aggregation is also referred to as implicit joins, for reasons that become obvious by reading the discussion on virtual relations. However, these implicit joins are not what Applicants refer to as implicit joins in the present invention. In these teachings, implicit join refers to the equi-join constraint that is generated based on the output-dimension in SDL set definitions. Although related, this is different from the implicit join that results from path expressions.

A relatively recent comparison of relational query languages (RQL) and object query languages (OQL) by Brown, S. "The Semantics of Database Query Languages", PhD dissertation, University of Sheffield, UK, 1999, points out important semantic differences between OQL and RQL in "grouped" queries. Although SDL has some resemblance with OQL (virtual dimensions), it is strictly connected to the relational model and mathematical aggregate operators behave identical in SDL as SQL. SDL is thus closer to object-relational extensions of SQL (SQL3). For the equivalence of the output-dimension in SDL, both SQL and OQL require the definition of a cursor, thus those languages don't support implicit join, in the meaning P:\OPERMHUI071423 sp I.doc-15101~00 00 of the word in the present invention, although they support de-referencing and path expressions.

Another technical point that emphasizes the difference between SDL and t OQL/SQL is that in SDL multiple virtual dimensions can be combined in a recordoperator, thus constraints on multiple attributes is possible in SDL even though "path Sexpressions" are used. It is impossible to use path expressions in OQL/SQL to refer to multiple attributes simultaneously without introducing an intermediate cursor or the use of Ssuccinct notation (see M. Stonebraker. "Object-relational DBMSs: the next great wave", Morgan Kaufmann Publishers, Inc. 1996).

Accordingly, there are other differences between SDL and both OQL and SQL, both in terms of their language structure as well as the underlying data models.

SUMMARY OF THE INVENTION In accordance with the invention, there is provided a method of retrieving sets of object data from a data store of a computer system by defining queries on virtual relations which define desired sets of object data to be retrieved from the data store, comprising the steps of: providing a written representation of a desired object data set in terms of dimensions from relations storing object data in the data store; implying constraints on relation instances or dimensions including specifying sets of objects using constraints on dimensions and using a record operator to indicate which constraints apply to a same tuple in a relation, said step of implying constraints enabling length of the written representation to be minimized; and using the written representation, querying the data store and retrieving the desired object data set, including using the record operator to apply multiple atomic constraints on different dimensions in a same tuple in a relation, said querying and retrieving resulting in the desired object data set as query results.

In another aspect, there is provided an apparatus for retrieving sets of object data from a data store of a computer system by defining queries on virtual relations which define desired sets of object data to be retrieved from the data store, comprising: an input component for providing a written representation of a desired object data P:OPERDIf0I7I23 s paI dc-ISOVl2a00 00 -6t set in terms of dimension attributes from relations storing object data in the data store, Swherein the written representation specifies sets of objects using predicates on dimension attributes and employing a record operator notation to indicate which predicates apply to a t same tuple; and ¢€3 an assembly coupled to receive the written representation, in response the assembly C)using the written representation to query the data store and retrieve the desired object data set by implying constraints on relations and determining number of dimensions and tuples Sin a relation to apply predicates, said implying constraints enabling length of the written representation to be minimized, the assembly producing an indication of the retrieved desired object data set as query results.

The written representation may include any combination of a disjunctive expression and a conjunctive expression. The assembly can thus perform OR-distribution on disjunctive expressions and eliminates from disjunctive expressions, conjuncts with undefined binding variables. The assembly translates conjunctive expressions to respective SQL join terms and translates disjunctive expressions to respective SQL-union terms. Further, the assembly rewrites the disjunctive and/or conjunctive expressions such that the SQL union operator is applied after the SQL join terms are calculated. This results in a computationally faster implementation.

Preferably, the assembly automatically enforces a record-operator where an expression in the written representation without the record-operator is equivalent to the expression with the record-operator or based on information on data structures of dimensions.

The written representation may include an IN-statement and a disjunctive expression in a nested set. In that case, the assembly can apply OR-distribution within the nested set by treating the IN-statement effectively as a record-operator expression.

In one embodiment the data store has a native query engine. In turn, the assembly further translates the written representation into code for the native query engine in a manner such that the code is optimized for querying the data store.

In one embodiment, the written representation utilizes a certain symbol, such as a colon, to specify hierarchical constraints on dimensions.

The written representation may include any combination of AND and OR P:AOPERDH 0171423 Spaldcx-I/01/200 00 -7expressions. In response, the assembly optionally performs OR-distribution in a manner that results in expressions with different sets of dimensions (dimension fingerprint). That is, if the conjuncts (AND expressions) inside a subject written representation have different V)sets of dimensions, the written representation is separated into multiple working expressions.

SThe written representation may include aggregate operators in expressions. Then assembly automatically introduces an implicit grouping based on the output-dimension and Splaces constraints based on the aggregates into a HAVING clause in SQL.

Preferably, syntactic gates are defined that allow SDL to be combined with SQL, thereby allowing bidirectional inlining of the two languages. Another important step to achieve this is to introduce operators to map the SDL metadata to the SQL metadata, e.g.

table names and column names. In one embodiment the assembly uses a double-recursive pass-through approach for compiling statements that include both SQL and SDL, thereby ignoring most of the complexity associated with maintaining a formal SQL parser.

In one embodiment, the input component includes an editor for composing written representations and (ii) a search engine for enabling user browsing of metadata, dimension values and relations of the data store. The search engine assists a user in composing desired written representations. In particular, the search engine provides graphical views of dimension hierarchies for user browsing. Further, the editor employs a user interface that supports drag and drop of dimensions and relation values in written representations being composed. Another aspect of the editor is to provide a syntactic and semantic feedback, based on the SDL syntax and the available domain definitions. Yet another aspect of the editor is to provide a descriptive representation of the SDL expression based metadata descriptions of the dimensions and the domain values.

Preferably, the assembly enables virtual dimensions and corresponding virtual relations to be defined based on domains primary dimensions. Virtual dimensions minimize metadata maintenance and allow attributes from different objects to be used together in an intuitive fashion, thereby hiding joins from the user. The search engine for browsing the metadata also utilizes information on primary dimensions for nested dimension drilling and thereby supports the generation of virtual dimensions in a drag-anddrop fashion. Alternatively, it also supports the construction of nested SDL queries with P:\OPERUDH10171423 spaldoc-ISmll200S 00 -8nested dimension drilling.

In one embodiment, the input component includes a dialog builder that constructs dialogs that can issue SDL queries with dimension attributes defined by the users in a dragt and-drop fashion. The dialogs have built in logic such that they generate an SDL expression based on only the dimension attributes that are used by the user. Thus the SDL Sgeneration mechanism of the dialogs eliminates unknown dimensions from the SDL expression.

Further, the preferred embodiment presents two mechanisms to federate SDL systems, namely, static federation and dynamic federation. For query evaluation performance the system relies on a federated RDBMS optimizer (see L.M. Haas, et.al., "Data Integration through Database Federation", IBM Systems Journal Vol. 41(4), 2002), however, the assembly provides a mechanism to map dimension names with prefixes as well as a mechanism to define dimension aliases. Thus metadata from multiple SDL systems can be combined in a relatively simple manner. A more unique federation mechanism in the present invention is the dynamic federation. This approach relies on the same metadata transformations, however, each user session can have different metadata composition from multiple metadata schemas specified at runtime. With dynamic federation, user and session dependent application data can be combined in SDL queries with other global server based data.

In another embodiment, the written representation has an expression with predicates on multiple attributes per conjunct in a non-ambiguous manner. The assembly enforces the expression using automatic record locking or a minimum cursor principle. When automatic record locking, the predicates on attributes from the same relation (CDC) are automatically enforced on the same record (or cursor). Also the assembly may enforce non-ambiguous expressions with multiple concrete dimensions or virtual dimensions or both, using automatic record-locking based on the minimum cursor principle. Where the expression has multiple references to a same attribute, the assembly enforces the expression using explicit record-locking based on the minimum cursor principle.

In another embodiment, the written representation has an expression with an extended virtual relation. The extended virtual relation may include predicates on dimensions and/or (ii) a WHERE clause within a record operator. The expression may employ nested record-operators. Where the extended virtual relation has dimensions from more than one relation (CDC), the assembly transparently assembles the virtual relations with equi-joins or outer-joins of the primary dimensions, based on the predicates in the extended virtual relation.

The extended virtual relation can contain virtual dimensions, and the assembly transparently generates and assembles the virtual relations based on the minimum cursor principle and the path expressions of the corresponding virtual dimensions.

The extended virtual relations are used as relations in SQL in one embodiment. In other embodiments, the extended virtual relation includes aggregate operators, and the invention method automatically applies SQL GROUP BY conditions and predicates within each CDC, based on the dimensions listed in the output of the given extended virtual relation or the dimension within a nested record operator.

BRIEF DESCRIPTION OF THE DRAWINGS The foregoing and other objects, features and advantages of the invention will be apparent from the following more particular description of preferred embodiments I. Sof the invention, as illustrated in the accompanying drawings in which like referen _characters refer to the same parts throughout the different views. The drawings ar Snot necessarily to scale, emphasis instead being placed upon illustrating the princi of the invention.

Figure 1 is a block diagram illustrating an exemplary setup required to Simplement the preferred embodiment of the present invention. As shown, the syste consists of hard disks, a computer with RDBMS, a computer with the SDL server system, and a computer running the SDL client modules.

C, Figure 2 is a block diagram illustrating how the SDL server takes a set definition as an input and returns as output the elements that belong to the defined Figure 3 is an example showing two relations with dimensions and their corresponding domains. For the enumerable domains GENDER and DATE, their corresponding attribute tables are shown. The values in the DATE domain have a hierarchy structure.

Figure 4 is an illustration of how the SDL meta-data tables would represen scenario in Fig. 3.

Figure 5 is an example of how virtual relations are generated as views on tf underlying data in the repository based on the SDL query.

Figure 6 is an example of data denormalization and corresponding metadat Figure 7 is a block diagram describing the process for creating the virtual relations that are needed to evaluate each expression.

Figure 8 is a block diagram describing the process for translating a conjun sub-expression with one or more negations.

Figure 9 is a simplified diagram that describes the process for translating disjunctive and conjunctive expressions.

Figure 10 is an overview of parser steps in one embodiment of the optimization and translation procedure from SDL to SQL.

Figure 11 is an overview of the view-union optimization.

Figure 12 is an overview of automatic record-operator enforcement.

Figure 13 is a block diagram describing additional steps required for parsi SDL statements with SQL/SSDL inlining.

-11- SFigure 14 is a block diagram describing the process of parsing SSDL expressions (SQL/SSDL statements inlined with SDL) and generating the SQL code.

Figure 15 is an overview of the default dimension logic.

Figure 16 is an overview of the virtual dimension and virtual relation logic.

5 Figure 17 illustrates a computer architecture for static SDL federation.

SFigure 18 show interactions and schematic overview in dynamic SDL N federation.

Figures 19A and B are a composite data explorer and query composer.

Figure 20A and B illustrate relation detection assistance in the composite query-tool.

Figures 21A and B illustrate SDL keyword and dimension assistance.

Figures 22A and B illustrate SDL queries and set analysis using a Venn-tool.

Figures 23A and B illustrate output-specification for a report on the sets.

Figures 24A and B illustrate further the output-specification of Figure 23.

Figures 25A and B illustrate SDL query dialogs.

Figures 26A and B illustrate URL templates in SDL reports Figure 27 illustrates nested dimension drilling, nested sets, virtual dimensions, default dimension prefix and default output-dimension DETAILED DESCRIPTION OF THE INVENTION The following description of the preferred embodiment is to be understood as only one of many possible embodiments allowed by the scope of the present invention. Reference is made to the accompanying figures that form a part hereof.

Overview A complete SDL system setup is a combination of client and server modules that can be utilized to build software applications that allow users to evaluate arbitrary SDL queries in an easy manner. The SDL invention described here covers both the server part of the system as well as the software components to facilitate the creation of SDL queries, either with a specific SDL editor or through specialized SDL query dialogs. Applicants also briefly describe utilities for importing data and for the maintenance of the SDL meta-data.

-12- In order to understand the importance of a simple, yet powerful set definition language, consider the case where a computer user is searching for files in the operating system that have some properties. Although not thought of as such by most users, this use case is an example of a pattern where a user is defining a set of objects, e.g. files. Also, consider the scenarios where a user needs to define sets of affected individuals, for genetic linkage analysis, based on medical events such as diagnosis, treatments and measurements. Also, consider the selection of a set of genes that have certain properties represented with attributes associated to them. Yet another example could be the selection of genetic markers to be used for genetic analysis or the selection of archived linkage runs, stored in a database with relevant search attributes associated to them. All these use cases are examples of where the definition of a "set" is in common.

A very important design objective with the SDL language was to create a language that could be implemented efficiently, both with respect to query response time and data volumes. Indeed, the preferred embodiment of the SDL query system is implemented in such a way that SDL statements are parsed and translated into SQL statements and evaluated in a RDBMS. Not only does this simplify the SDL compilation and evaluation logic, but also it allows one to utilize the enormous effort that has gone into optimizing queries and the manipulation of large data volumes in RDBMS. Similar approaches have been used before in other language implementations, e.g. in LDAP and XML query language applications.

As shown in Fig. 1, The SDL server system 100 consists of several components 101. In the preferred embodiment the components 101 include a parser 11, optimizer 13, translator 1 5 (preferably SDL2SQL translator), evaluator and data and meta-data modules 17. In the preferred embodiment, the SDL server is also comprised of an RDBMS (Related Database Management System) 102 and hard disks 103 for the storage of the data. It is a matter of configuration whether RDBMS 102 and components 101 reside in the same computer or whether they are kept on different computers. The server 100 is then connected to. an SDL client 104 through a wide area or similar network 105. The client 104 can either be an application specifically -13designed for SDL or a SDL query component bundled into a host application.

Generally client 104 is formed of a query composer 19 and a search engine 21 of sorts.

The rest of the text describes the nature of the SDL language, logical constructs such as dimensions and domains, relational schemas as well as mapping of the language to the SQL language as well as bidirectional integration of SDL with SQL. Further, later described are special SDL query tools that facilitate the construction of SDL expressions.

Introduction to SDL The best way to understand the SDL language is by taking many short examples. Along the way Applicants introduce the concepts necessary for understanding the composition of a general SDL query. A general SDL query declaration looks almost like the way sets are defined with standard mathematical notation: setname output dimension SDL expression Where, "setname" is the name of the set that is being defined and "SDL expression" is an expression of first order logic that has to be true for every element in the set. The set is defined over the output dimension. The SDL language is in many ways similar to relational calculus (see R. Ramakrishnan and J. Gehrke, "Database Management Systems," 2nd ed., McGraw Hill, 2000), especially domain relational calculus (DRC).

Both of these languages are for instance unsafe as defined by Ramakrishnan and Gehrke. The main difference is that in the SDL language, dimensions are used as compared to domain variables in DRC. Therefore referrals to relations in SDL are unnecessary. Also, in SDL the output dimension is always implicit in the SDL expression, i.e. other dimensions have to appear in relations with the output dimension and a natural join on the output dimension is implemented behind the scenes. This results in very sparse expressions that are easily human readable. Typically the output dimension is an identifier of some objects that can then later be used to retrieve additional information on those objects that were in the set that was being defined.

Figure 2 shows how an SDL statement is sent to an SDL server 200 from client 104. The statement is comprised of the output-dimension D 201 and an SDL expression 202. The server 200 returns a set 203 with elements xi from the output- 14dimension xi eD). For each element xi there must exist relation instances in the data with the output-dimension D and other dimensions referred to in input expression 202 such that the expression in 202 is true for each element xi in the output set 203.

An example that brings this into clinical context is the following: patients pid I dob>1966 AND diag:icd9.stroke AND date>2000 Here, patients become the set of all personal identification numbers (pid) of individuals who were born after 1966 and have been diagnosed with a stroke after the year 2000. The proper way to read this example set definition is that "patients defines the set of all individuals for which there exists a date of birth attribute larger than 1966 and for which there exists a data record/relation with the attributes diag:icd9.stroke and a date larger than 2000." In addition to the output dimension of pid, the dimensions that appear in this definition are dob, diag and date. Note the use of the colon in "diag:icd9.stroke" which is equivalent to "diag=:icd9.stroke." This syntax is allowed for dimensions such as diag that are of a domain type that is enumerable and has a hierarchy associated with it. For such dimensions, this is a short hand notation for the SQL constraint "diag LIKE 'icd9.stroke.%' OR diag 'icd9.stroke' i.e. all diagnoses that are children or lower level elements in the corresponding diagnostic code hierarchy. The square brackets denote the socalled record operator, used to enforce the co existence of one or more attributes in the same record/relation. Since the same individual can have multiple diagnoses it is not uncommon to record them in relation with a date. However, certain attributes, such as date of birth, are each a singleton by nature and therefore typically only registered once per individual.

Another SDL expression example from the biology field is as follows: target genes gene id geneclass:gpcr AND expr.tissue:brain $x expr.level AND expr.tissue:stomach AND 1.5*expr.level $x Based on this definition, the set "target-genes" will contain all gene ids of GPCRs that are expressed 50% more in brain tissue than in stomach tissue. Note the use of binding variables within the record expressions to enforce the expression level in some brain tissue to be 50% higher than any expression level in stomach tissue. One sees that the output dimension, i.e. the identifier of the elements that represent the sets, does not appear in the SDL expression itself. Once the set type has been specified, it is implicit in the SDL expression that all of the other dimensions used refer to attributes associated with the output dimension. This is one of the unique features of the SDL syntax and results in very short expressions. In the first example, the existence of the relations (pid, dob) and (pid, diag, date) is assumed, and in the second example the existence of relations (gene-id, class) and (gene-id, expr.tissue, expr.level) is assumed.

The SDL system can be configured in such a manner that the output dimension has been preset, i.e. the SDL server 100, 200 returns only a single type of set, for instance sets of individuals, genes, markers etc. Thus, the user usually only has to create the "SDL expression" part itself, and does not have to specify the "output dimension" and the curly brackets in the SDL query; and the SDL query tool will generate an output set of the appropriate type. However, an SDL system can also be set up in such a manner that it allows definition of various types of objects.

Furthermore, it is within the purview of those skilled in the field to extend the invention language (SDL) in such a manner that it allows the definition of relation with multiple output dimensions. In this embodiment however, the applicants choose to integrate SDL with SQL for working with output data of relations as explained in later sections.

Domains, dimensions and relations Without going into a formal syntax specification, Applicants next briefly consider the structure of a general SDL expression. Without a loss of generality, assume an SDL system 100 (Fig. I) is configured to generate sets of individuals (pid s) based on longitudinal medical event data registered on those individuals. The general format of patient data imported into the SDL system 100 is therefore of the following relational form: relation (output dim diml, dimN)=(pid, For data to be directly applicable to set definition in SDL, the output dimension has to appear in the data relations (here it is pid). The other elements -16- (representing respective columns in a data store 103) that appear in the relation statement and represent one or more attributes associated with an individual, have to be of pre- specified dimensions. Thus, dimensions can be considered as attributes or terms associated with individuals through data relations. If data is normalized in such a way that the output dimension is not present, it can still be used through the use of nested sets. Nested sets and the registration of normalized data are discussed later in more detail.

In the present invention, instantiations of domains are referred to as dimensions. Thus, for each dimension, a domain has to be specified. Not only does the domain specify the data type used to represent its corresponding dimensions in the RDBMS 102, but it also specifies the logical content and constraints associated with its corresponding dimensions. For instance, the domain may specify whether the corresponding dimension is "closed" (enumerable) or "open" and whether there are maximum and minimum limitations. As an example, consider attributes such as weight and height. Both are naturally represented by a NUMBER, however, their values represent totally different physical measures. Therefore, two separate domains should be used to represent values of weight and height, and two dimensions from the different domains should not be comparable. Hence, the domains can be used to enforce strict type checking. As another example consider the dimensions "date of birth" and "date of death". Although these dimensions represent two different attributes, they have still the same logical type and are therefore naturally two instances of the same domain, i.e. date. In summary, dimensions represent attributes that are instances of a certain domain, and dimensions that are instances of the same domains are comparable. When a domain is defined as closed, its values (corresponding dimensions) can optionally be organized into a hierarchy (DAG, parent child relationship).

Dimensions must have distinct names. Their names can be organized in a hierarchical manner (folders) such that the path name represents the distinct name of the dimension. In such case, they are typically placed in the hierarchy based on their logical meaning. A GUI metadata browser that shows the available dimensions to a user would then typically show the dimensions listed in a tree based on their names.

-17- SHowever, the names of the dimensions do not have to reflect their position in the tree _of the metadata browser and the same dimension could even appear in multiple o locations. As a systematic approach, in order to register a relational schema with multiple tables into the SDL system 100, the dimensions could be named following the format "table name.domain". However, dimensions that represent ids (identifiers) Sof objects that are supposed to be in the SDL sets have to map with the same name for N each table.

SDimensions in the present invention are classified into several categories: PD: Primary dimensions; they are equivalent to object-identifiers (OLD) in 00 systems and used as output and domain specification in SDL set-definitions.

Only enumerable domains may have PDs associated with them. Primarykeys (PK) and foreign-keys (FK) in RDBMS 102 are typically mapped with corresponding primary dimension.

RD: Reference dimensions; they are equivalent to object-references. Each RD is of the same domain as a corresponding PD. RDs are typically mapped to FK in RDBMS 102.

AD: Attribute dimensions; these are dimensions that are neither PD nor RD but used to qualify objects.

Concrete relations. Concrete relations are defined in SDL as an unordered collection of two or more dimensions that are registered with the system. Each relation must have one and only one PD and each dimension can only appear once in each relation. When a concrete relation is defined, it is specified if the PD has a distinct constraint in the relation (primary key PK).

A dimension name represents the role that the corresponding domain plays in the given relation, however, relation names in SDL do not have any particular function but rather the so-called dimension fingerprint, i.e. the list of dimensions that defines the relation.

Virtual relations. By definition, a virtual relation (VR) is a relation that is defined by a dimension fingerprint. The virtual relation is the main data abstraction in SDL and -18used to hide or encapsulate the storage implementation of the concrete data relations stored in the RDBMS 102. Based on how VR are implemented, they can also be used to provide access to only selected parts of data, based on scoping parameters such as user privileges etc.

Relation overlap: Two concrete relations are said to overlap if they have in common dimensions other than their primary dimension. Formally, two relations RI,R 2 e Z, where E is a superset that denotes the data schema, are said to overlap if and only if: dp e R I A d E R 2 A 3d a R I {dp} Ad e R2.,where dp represents their primary dimension. Applicants denote the overlap of two relations R, and R 2 with the Boolean function O(RI,R 2 CRC and CDC: A concrete relation cluster (CRC) is a set of concrete relations in the schema, Y, that through transitive closure can be linked through overlapping concrete relations in I. Formally, a set of concrete relations C c Y is a concrete relation cluster if and only if:

VRI,,R

2 e C,3S r 2 r,)A O(rl,r2)A...A(r R 2 Likewise, a concrete dimension cluster (CDC) is a set of dimensions that form the relations in a CRC.

Virtual relation: Given a dimension fingerprint F that is a CDC with a primary dimension dp, the corresponding virtual relation is defined as 7r

F

VR, e where dp E For concrete relations where 3d E F A d e R, the dimension values are substituted with "missing value" (NULL) in their tuples.

As mentioned above, there can be only one PD in each concrete relation and the same holds for virtual relations. Overlapping concrete relations must have the same constraint on the PD and the PD is assumed to be unique in the corresponding CDC if the relations have PK on the PD. Whether and how such uniqueness is ensured is implementation dependent.

-19- Collections and singletons: Singletons are dimensions in a CDC in which the PD has a unique constraint Correspondingly, dimensions in a CDC that are not singletons are defined as collections.

Figure 3 shows an example of two relations, 300 and 301, and four different domains, 303, 304, 305 and 302. Relation 300 has six dimensions, namely CPID, CFID, CMID, CSEX, CDOB and CDOD. Relation 301 has three dimensions CPID, CDIAG and COBS DATE. Two of the domains 303, 304 (diagnosis and gender, respectively) in this example are enumerable and therefore have associated with them tables 306, 307 that list the possible values in the domains, 304 and 303. Diagnosis Domain 303 table 307 has a hierarchical nature, however, gender domain 304 table 306 is a special case of a flat diagnosis hierarchy. The other domains, identifier domain 305 and date domain 302, are open domains that do not have tables that list the possible values associated with them. Both relations 300 and 301 have the dimension PID (in column CPID). Also, relation 300 has three different dimensions (CPID, CFID and CMID) that belong to the same domain, i.e. Identifier Domain 305.

It is worth emphasizing that there is an important difference between dimensions of the invention SDL and conventional RDBMS columns. Dimensions can be mapped to multiple different columns, as long as they belong to different tables or views. Likewise, the same column can be mapped with multiple SDL dimensions and a table or view can be mapped multiple times, for relations with different PDs.

Similarly, there is a difference between VR and regular tables or views in RDBMS.

Virtual relations do not have names as such but are dynamic relations identified from their dimension fingerprint and collected at runtime.

Tables and metadata Available relations in an SDL repository are completely defined by the data that has been imported or registered with the system. A relation is very much like an SQL database table but the columns do not bear names as such, rather each column is bound to a specific dimension. Each dimension may appear in an arbitrary number of relations. The dimension may be thought of as a super column able to span an arbitrary number of tables. The SDL language defines sets of objects that depend on the output dimension in the set definition based on relations registered with the SDL system 100. The SDL language neither specifies how expressions are evaluated nor the storage mechanism for the relations. When RDBMS 102 (Fig. 1) is used for the embodiment of the system, two extreme data schemas can be used for the storage of relations. Both of these schemes set no limits to the number of relations allowed. One approach is the so-called horizontal structure, i.e. a large table with sufficient number of columns to store all the combinations of dimensions that can coexist together. The other extreme alternative is the vertical representation or a fully pivoted storage format (see R. Agrawal et al., "Storage and Querying of E-Commerce Data," Proceedings of the 27th VLDB Conference, Rome Italy, 2001) that has recently been proposed as a storage mechanism for data where there are very many different logical relations. In the vertical schema, all relations are put into a single table that has only three columns, i.e. rowid, dimension, and value.

Other alternatives have also been proposed for schemas that are allowed to evolve, such as to store the relations in multiple 2 ary tables (see S. Shi et al, "An enterprise directory solution with DB2", IBM Systems Journal, 39(2): 360 383, 2000; M. Missikoff, "A domain based internal schema for relational database machines", In Proceedings of the 1982 ACM SIGMOD International Conference on Management of Data, Orlando, Florida, June 2 4, 1982, pg. 215 224; G. P. Copeland and S. N.

Khoshafian, "A decomposition storage model", In Proceedings of the 1985 ACM SIGMOD International Conference on Management of Data, Austin, Texas, May 28 31, 1985, pg. 268 279; and S. Khoshafian et al., "A query processing strategy for the decomposed storage model", In Proceedings of the Third International Conference on Data Engineering, February 3 5, 1987, Los Angeles, California, USA, pg. 636 643).

The present invention proposes a new alternative that is closer to regular relational schemas used in most RDBMS. Applicants' approach is to store information on all the dimensions that have been defined in the system 100, their domain and the relations they exist in. The preferred embodiment defines the following metadata tables (at 17 in Fig. 1) to store information on domains, dimensions and relations: -21 domains(domain, SQLtype, SQLattributehierarchy) dim2dom(dimension, domain) relations(relation, SQLrelation, inclusion_criteria) dim2rel(dimension, relation, column_name, multiplicity) These four SQL tables define a basic metadata structure for the invention SDL server system 100. One embodiment omits data types on the columns in these tables and effectively assumes column contents to be strings VARCHAR2 in Oracle

RDBMS).

The table named "domains" stores a domain name and the SQL data type used for columns belonging to dimensions defined on its domain. For domains that are enumerable and closed, it also stores reference to a table listing all the allowed values in a hierarchical manner (the hierarchy can also be flat). Optionally, minimum and maximum values can be used to specify an allowed range for open domains.

The table "dim2dom" stores indications of the connections between dimensions and their corresponding domain. The table named "relations" indicates connections between SDL relation and corresponding SQL structure for representing that relation. The SQL structure can either be a table, view or materialized view. For each relation, an inclusion criteria can be specified, i.e. a condition on the selected dimensions that needs to be met for the table to be included into the SQL code (see the following discussion). Formally, Applicants' use of relation is more like a set of relation instances since the same SDL relation can exist in more than one SQL relation. This will become clearer later.

Finally, the table "dim2rel" indicates the connection between the dimensions and the relations (set of relation instances) and indicates the column used to store the dimension in the corresponding table. The last column stores information on whether duplicates are allowed in the dimension in the corresponding relation.

Figure 4 shows how the meta-data tables 17 mentioned above, would be instantiated for the example relations 301, 302 of Fig. 3. In that example, domains table 400 specifies the four domains (Diagnosis 303, gender 304, date 302 and identifier 305) and their respective SQL type (string, character, date, integer, respectfully). Domains table 400 makes reference (Tdiags, Tsex) to enumeration -22tables 306, 307 for the closed domains 303, 304 and indicates maximum and/or minimum values for the open domains 302, 305. Dim2dom table 401 indicates the domain. As further explained below, Table 403 shows dimensions cross-referenced to relations (dim2rel) and Table 402 provides correspondence between an SDL relation and a respective SQL structure.

Referring back to Fig. I, the SDL server 100 uses its metadata tables (collectively illustrated at 17) to parse (at parser 11) an SDL expression and translate it (via translator 15) into a corresponding SQL statement that refers to the appropriate tables that are needed for proper evaluation of the subject expression. This is best explained through a short example.

Consider the following SDL expression: dO I dl 2 Here dO is the output dimension and the constraint is set on dimension dl. This expression should return all values of dO that exist in relation with dl where d l 2. Expressed in relational algebra, the following is equivalent: dO I dl 2 ndO(Od=2((dO,dl))) To evaluate this expression, the invention SDL system 100 uses the table dim2rel (like 403 of Fig. 4) to look up all relations that contain the dimensions dO and all the relations that contain the dimension dl. Two sets of relations result, one for each of dO and dl. Next the invention SDL system 100 takes the intersection of these two sets of relations. The set resulting from the intersection can contain one or more relation name. The system 100 then creates a virtual relation (VR) that is a projection of the two dimensions in the union of all the relations it found with the intersection operation union of sets of relations instances). This virtual relation is denoted "v01", the numbers representing the dimensions that exist in it.

Let's assume that there are only three tables registered with the SDL server 100 as shown in Fig. 5: Tablel (dO,dl) 500, Table2(dO,d2) 501, and Table3(dO,dl,d2) 502.

For simplicity of the example, assume that the column names of the tables are the same as the name of the dimensions they store. This is not a requirement, it is just for purposes of illustration in the example. The SQL code that evaluates the previous SDL expression is then: -23- SELECT DISTINCT DO FROM (SELECT V01.DO DO FROM (SELECT DO, D1 FROM TABLE1 UNION SELECT DO, D1 FROM TABLE3) V01 WHERE V01.D1 2); Notice how the virtual relation Table v01 503 only contains tablel and table3 since table2 does not contain dimension dl. An alternative way to define the virtual relations on the fly in an SQL statement is by using the WITH statement and common table expression. The WITH statement however only exists in some new versions of SQL implementations, such as IBM DB2 and Oracle9i. Similar to using the WITH statement, with respect to an SQL syntax virtual relation v01 of Table 503 could also have been defined as a view, instead of being defined on the fly as above: CREATE VIEW V01 (DO,Dl) AS (SELECT DO, D1 FROM TABLE1 UNION SELECT DO, D1 FROM TABLE3); SELECT DISTINCT DO FROM (SELECT V01.DO DO FROM V01 WHERE V01.D1 2); If the registered data consists of the tables 500, 501, and 502, the result from both of these SQL statements shown above would be the set f} 504.

Consider another more complicated example with a record operator: dO dl 2 AND d3 0 AND d4 0 7tdO(Tdo=dO' (Cdl=2 (dO,dl) x Od3>O,d4<o ,d3,d4) In order to evaluate this expression, the SDL system 100 creates two virtual relations, vO1 and v034 and performs a natural join on dO (the prime is just used to make a distinction between the same dimension coming from two different relations).

In the following discussion on translation of SDL to SQL statements, the virtual relations are denoted with and a suffix based on the dimensions that are included in the corresponding relation. For instance, a relation with dO, dl and d3 is denoted by v013. The following will leave aside whether the virtual relation is defined on the fly or with SQL views since it does not impact other things in the SQL statement structure.

-24- Normalization and views Schema refinement through normalization of data is a common practice in relational schema designs and there exists extensive literature on this subject (see R.

Agrawal et al., "Storage and Querying of E Commerce Data", Proceedings of the 27th VLDB Conference, Rome, Italy, 2001 417)). The basic idea is to format the data such that updates are easy, the data is flexible with respect to queries, and storage space is minimum. The invention SDL system 100 utilizes normalized data in two ways: directly, through the use of nested sets, and indirectly through the use of views.

Here a closer look at the former approach is taken since, due to limitations in most SQL optimizers, it requires a special flag in the metadata in order to be implemented efficiently.

Without loss of generality, consider an arbitrary scenario where one has two relations as shown in Fig. 6, diags(pid, diag, hospid, docid, date) 600 and hospitals(hospid, name, type, region, zip) 601. If one would like to define a set of all patients that have gotten some diagnosis in a hospital in a particular region, the corresponding invention SDL expression could look like: pid I [diag icd9.x AND hospid IN hospid I zip 101 This expression is perfectly fine, however, it can be argued that it is relatively long and complex for most inexperienced users. One might like to define new dimensions on the diagnoses that make this definition easier. One could for instance create a view 602 called vdiags and register it in addition to diags 600 and hospitals 601: vdiags(pid,diag,hosp.name,hosp.type,hosp.region,hosp.zip,doc.speciality,doc.age,date with the following SQL code: CREATE VIEW VDIAGS AS (SELECT

PID,DIAG,HOSP.NAME,HOSP.TYPE,HOSP.REGION,HOSP.ZIP,DATE

FROM DIAGS, HOSPITALS HOSP WHERE DIAGS.HOSPID

HOPS.HOSPID);

With these dimensions, the subject SDL expression would simply become: pid [diag icd9.x AND hosp.zip 101] This expression is obviously much shorter and easier to understand than the one above (hosp.zip could also have been called diag.zip).

There is a dangerous pitfall however. Notice what happens if one creates an expression such as: pid I diag icd9.x As discussed above, this expression will generate a virtual relation that is a union of diags 600 and vdiags 602, since both of these tables contain the dimensions pid and diag. All the information on the relation (pid, diag) is in diags 600, and vdiags 602, adds no additional information given its definition. What is even worse is that the SQL code that would be generated (see the following discussion on the translation of SDL to SQL) would use the view vdiags 602 with no conditions on the dimensions taken from hospitals 601. Hence, this would result in a cross product (Cartesian product) between the tuples in diags, 600 and hospitals 601, a join that could potentially be very expensive. This is because SQL optimizers cannot recognize that this join does not have to be performed, given the output parameters and the conditions specified in the SQL join.

Applicants' solution to the above problem is not to abandon views, but to allow them to be registered into the SDL system 100 with additional information on their inclusion criteria. The inclusion criteria can be implemented in many ways. For instance as a list of sets of dimensions with the meaning that one dimension from each set is required to appear in an expression or record expression (see the syntax specification) for the table/relation to be included into the virtual relation. As an example, Applicants could have specified the inclusion criteria 604 on vidags in the metadata table relations 603 as: {pid, diag, hospid, docid, date}, {hosp.name, hosp.type, hosp.region, hosp.zip}. Then the view would only be included into the virtual view when there is at least one column required from each of the two tables, diags 600 and hospitals 601. A query that uses dimensions from both of the underlying tables, diags 600 and hospitals 601, would however only use the view 602, vdiags, since the inclusion criteria would be false for the two tables.

In general, the inclusion criteria will result in fewer SQL relations that will be included into the SQL statement. Figure 7 summarizes the process of finding which -26- SQL relations are needed in a virtual relation for an SQL statement that results from the translation of SDL to SQL. An SDL statement is provided as input. At step 701, the system sets the output dimension as the implicit dimension from the input statement. An output dimension inside a nested set definition overrides the previous output declaration.

Next in step 702, for each sub expression in a conjunct, record-expression, relational-expression, or a simple expression, the system determines and collects all the distinct dimensions that are referred to, into a set. The system includes the implicit dimensions.

In step 703, the system finds from the metadata and lists all the SQL relations that include the dimensions in the set created in step 702. The system eliminates from the list the relations where the set of dimensions does not meet the inclusion criteria.

In step 704, for each relation that was found in step 703 for each subexpression, the system applies a union operation to create a virtual relation. The system then uses the virtual relation in the corresponding SQL structure that evaluates the corresponding sub-expression.

It is worth mentioning that star-schemas, see William A. Giovinazzo, "Object- Oriented Data Warehouse Design: Building a Star Schema", Prentice Hall, (Feb.

2000) and Oracle 9i B "Data Warehousing Guide", (Part Number A90237-01), June 2001, Oracle Corporation B www.oracle.com, are common examples of where data is denormalized. A star usually refers to a fact-table with corresponding dimensiontables. The fact-table usually stores facts that are some kind of measures, e.g. cost, and dimension columns that are classifiers on the measure (attributes). The dimension-tables are then composed from a dimension column that has a foreign-key relationship with a dimension column in the fact-table in addition to more columns that are usually lower-resolution classification of the dimension, e.g. timestamp grouped into weeks, months, quarters and years. In the terminology presented herein, Applicants do not make any distinction between facts and dimensions as such, although, dimensions would typically be dimensions that belong to enumerable domains (and be primary dimensions, see later discussion on primary dimensions and default output-dimensions).

-27- Applicants have described that the SDL metadata 17 provides special support for enumerable domains that have hierarchical values. The SDL metadata 17 could also be augmented in such a manner that domains could be assigned dimension-tables, i.e. dimension grouping tables. This could for instance be obtained by adding fields in the domain table, see 400 in Fig. 4. The benefit of this would be that the SDL system could automatically provide the grouping that the dimensions-table provides to any dimension that belongs to the corresponding domain.

As an example, consider the import of registration of a relation that has a dimension with a date dimension that is an instance of the domain DATE. If a dimensions-table, dategr(date, week, month, qt, year), had been assigned to the DATE domain, the SDL system could automatically create, in addition to the date dimension, a dimension such as date.week, date.month, date.qt, date.year and register a view (or materialized view) that is a join between the imported relation and the dimensiontable. This would provide the user with a "hierarchical-like feeling" for all dimensions of such domains, when creating an expression with a constraint on time.

Translation of SDL to SQL To explain how a general SDL statement can be translated to a SQL statement, this discussion uses a stepwise approach and starts by explaining how the basic SDL structures map to SQL. For one skilled in the art of compilers and RDBMS, it will then follow how to construct a generic translator/compiler 15 that takes a general SDL statement that complies with the SDL syntax specification and maps it to an SQL statement. The SQL standard is defined by ANSI document, X3.135 1992, "Database Language SQL" and in revised form by document ANSI/ISO/EIS 9075. These documents are available from the American National Standards Institute.

Atomic expression Consider the most simplistic definition of an SDL set as: dO I dl Here dO is the output dimension, i.e. the set is defined on dimension dO. This SDL definition is translated into the following SQL: -28- SELECT DISTINCT DO FROM (SELECT V01.DO DO FROM V01 WHERE V01.D1 NULL); where v01 is defined as a view of all relations with dimensions dO and dl, i.e.

(dO, dl). Since dO is the output dimension, according to the SDL syntax, a specific (or target) dimension for the output does not have to be mentioned in the expression itself.

Constraints and comparisons SDL expressions with relational expression and calc-expr are translated in the following manner: dO dl pi is translated to: SELECT DISTINCT DO FROM (SELECT V01.DO DO FROM V01 WHERE V01.D1 PI); dO I dl pi is translated to: SELECT DISTINCT DO FROM (SELECT V01.DO DO FROM V01 WHERE V01.D1 PI); Two dimensions with the same domain type can be compared in the following manner: d0 dl d2 The above expression is translated to: SELECT DISTINCT DO FROM (SELECT V01.DO DO FROM V01, V02 WHERE V01.DO V02.DO AND VO1.D1 V02.D2); Notice that since no record operator (see below) is used dl and d2 do not have to come from the same tuples nor the same relation.

The same is the case in this calculated expression: dO dl d2 10 which is translated to: SELECT DISTINCT DO FROM (SELECT V01.DO DO FROM V01, V02 WHERE V01.DO V02.DO AND V01.D1 V02.D2 -29- Conjunctive expression dO dl pi AND d2 4 is translated to: SELECT DISTINCT DO FROM (SELECT V01.DO DO FROM V01, V02 WHERE V01.DO V02.DO AND V01.D1 PI AND V02.D2 4); Disjunctive expression dO dl pi OR d2 4 is translated to: SELECT DISTINCT DO FROM SELECT V01.DO DO FROM V01 WHERE V01.D1 PI UNION ALL SELECT VO2.DO DO FROM V02 WHERE VO2.D2 4) Record operator dO I dl pi AND d2 4 is translated to: SELECT DISTINCT DO FROM (SELECT V012.DO DO FROM V012 WHERE V012.D1 PI AND V012.D2 4); Like before, v012 defines a view of all relations with dimensions dO, dl, and d2. The previous SDL expression could also have been written in this way: dO I dl pi AND NOT d2 4 and it is translated to: SELECT DISTINCT DO FROM (SELECT V012.DO DO FROM V012 WHERE V012.D1 PI AND NOT V012.D2 4); Notice the equivalence of NOT and inside the record expression. The meaning of NOT outside record expressions is different as described next.

Consider now two examples that use relational expression and calc-expr inside the record operator and contrast them with the examples shown previously. Two dimensions with the same domain type can be compared in the following manner: do I dl d2 SThis is translated to: SELECT DISTINCT DO FROM (SELECT V012.DO DO FROM V012 WHERE V012.D1 V012.D2); Notice that since a record operator is used, dl and d2 have to come from the same n 5 tuple in a virtual relation. The same is the case in this calculated expression: dO I dl d2 10 which is translated to: SELECT DISTINCT DO FROM (SELECT V01.DO DO FROM V012 WHERE V012.D1 V012.D2); Figure 7 explains the logic behind the generation of virtual relations. The process receives on input an SDL expression. Step 701 sets the "current" outputdimension, as a required dimension in each virtual relation. In step 702, virtual dimensions that belong to the same record are found based on record-operators, calculated expressions etc. A distinct list of such dimensions, including the implicit dimension, determines the dimension fingerprint. Translation of each sub-expression conjunct may require several virtual relations with different fingerprints. Based on the SDL metadata, all SQL relations that fulfill the inclusion criteria for each fingerprint are found 703. In the present embodiment a virtual relation is generated for each dimension fingerprint 704 by using the SQL UNION keyword and common table expressions.

Negations Before considering the general case of expressions with NOT, start by considering two simple cases: dO dl pi AND NOT d2 4 is translated to: SELECT DISTINCT DO FROM SELECT V01.DO DO FROM V01 WHERE V01.D1 PI MINUS SELECT V02.DO DO FROM V02 WHERE V02.D2 -31- Here the SQL keyword MINUS is used to perform set-subtraction. The MINUS keyword is Oracle specific, however, MINUS can be replaced with EXCEPT in order to make the SDL translation SQL ANSII compliant.

Consider now the simpler case where there is only the NOT term in the expression: dO I NOT dl pi Now it seems impossible to use the set minus approach shown above. However, by recognizing that since dO is the output dimension, it is implicit in the constraint and therefore this set definition is equivalent to: dO I dO AND NOT dl pi and can therefore be translated to: SELECT DISTINCT DO FROM SELECT VO.DO DO FROM VO WHERE VO.DO NULL MINUS SELECT VO1.D DO FROM V01 WHERE VO1.DI PI H-lere vO denotes a view containing all relations with dO. Applicants like to refer to v0 as the "universe" or all the attributes that the dimension dO covers in its corresponding domain.

It should be mentioned that the inclusion-criteria, e.g. 604 in Fig. 6, on relations can be used to control which SQL-relations will be included in the virtual relation that constitutes the "universe". Also, the universe, vO, of a given outputdimension, dO, can also be implemented as a materialized view. Thus an SQL optimizer can substitute a union of possibly multiple tables with a summary of all the possible values for that dimension. Furthermore, as discussed in the section on primary dimensions and default output-dimension, a universe of a given outputdimension can be declared in terms of its domain and relations representing the complete listing of values in the corresponding domain.

Now consider a record expression with NOT: dO I d3=7 AND NOT dl pi AND NOT d2 4 The above is translated to: -32- SELECT DISTINCT DO FROM (SELECT V03.DO DO FROM V03 WHERE V03.D3 7 MINUS SELECT V012.DO DO FROM V012 WHERE V012.D1 PI AND NOT V012.D2 4) Like before, v012 defines a view of all relations with dimensions dO, dl, and d2.

Binding variables Binding variables can be used in conjunction with records, e.g.: dO I dl a $x d2 AND dl b AND d2 $x This definition is translated to: SELECT DISTINCT DO FROM (SELECT VA.DO DO FROM V012 VA, V012 VB WHERE VA.DO VB.DO AND VA.D1 a AND VB.D1 b AND VB.D2 VA.D2); Notice how the criteria with the binding variable are implemented using additional constraints on the join, which in this case is a self-join.

An alternative way for defining binding variables in the syntax can be the following: dO dl a AND $x d2 AND dl b AND d2 $x This way of declaring binding variables is easily implemented if it can be guaranteed that dimension names will not cause naming conflicts. This is most easily achieved by using a different assignment operator than the equality comparison operator, i.e.

Applicants use a Pascal like operator, for assignment. Furthermore, Applicants use the symbol in the beginning of binding variables, simply to make them more easily recognized in the teaching of this invention. In the SQL translation above, the translator replaced reference to $x with a reference to d2.

An alternative way is to rename the SQL-column that is created in the SQLimplementation of the virtual relation by using nested table expressions, e.g.: SELECT DISTINCT DO FROM (SELECT VA.DO DO FROM (SELECT V012.DO DO, V012.D2 BX FROM V012) VA, V012 VB WHERE VA.DO VB.DO AND VA.D1 A AND VB.D1 B AND VB.D2 VA.BX); -33- SHere, the relation VA contains a column named BX that defines the binding variable.

_This method makes it easy to extend the binding variable declaration such that formulas can be defined, e.g.: dO I [dl a AND S$x d2/dl AND dl b AND d2 SIt goes without saying that this translation strategy also applies when binding variables Sare declared with the other approach, i.e. behind a semicolon.

Yet another way of declaring binding is by using a syntax more related to SQL cursors, i.e. bind relations as compared to individual columns or functions. As an example consider the following: dO I dl a AND d2 as $x AND dl b AND d2 $x.d2 A natural way to translate this to SQL is: SELECT DISTINCT DO FROM (SELECT VA.DO DO FROM V012 VX, V012 VB WHERE VX.DO VB.DO AND VA.DI a AND VB.DI b AND VB.D2 VX.D2); Here VX becomes the name issued to the first virtual relation that has the dimension fingerprint 012, because of the binding declaration of the relation.

In all of these examples, Applicants have not been very rigorous with regard to the naming of the SQL columns, but in actual SDL2SQL implementation, it is important to have a naming system that avoids naming conflicts. The following examples will only use the first two approaches for defining binding variables, however, the translation logic discussed below should apply to all syntax variants.

Binding variable and negation Binding variables can be used in conjunction with records, e.g.: SdO I [dl a $x d2 AND NOT dl b AND d2 Because of the binding variable, it is no longer possible to use a straight-forward set minus approach for implementing NOT. Applicants therefore translate this in the following manner: -34- SELECT DISTINCT DO FROM (SELECT VO1.DO DO FROM VO12 VA WHERE VA.D1 a AND NOT V012.DO IN (SELECT VB.DO DO FROM V012 VB WHERE VA.DO VB.DO AND VB.D1 b AND VB.D2 VA.D2) Note that this approach could also have been used for implementing NOT where there are no binding variables, however, it is usually slower than the minus approach. Also notice the renaming of the relations because of the multiple referrals to the same relation. Finally, notice the constraint vA.dO vB.dO that is important for many optimizers when nested SQL is used with negation. Figure 8 summarizes the process for translating a conjunctive SDL expression with negation to SQL.

The process receives on input a conjunctive SDL expression. In response to the input, the process (at step 801) reorders the sub-expressions in the subject conjunct such that sub-expressions with negations succeed other expressions. The process also provides that sub-expressions with references to binding variables precede subexpressions without binding variable references.

In step 802, if all the sub-expressions have negation or there is just one subexpression and it has negation, then the process augments the expression with a subexpression that is the corresponding output dimension, without any additional constraint. Next step 802 redoes the reordering of step 801.

In step 803, if there is a reference to binding variables in the sub-expression, then the process uses the IN structure approach in the SQL translation. Otherwise, the process uses the MINUS structure.

Nested sets The invention SDL syntax allows for nested sets through the use of the IN keyword in expressions. Consider the following simple case: {dO dl 3 AND d2 IN d3 d4 3 Notice that for this definition to be valid the two dimensions, d2 and d3, have to be from the same domain. The most obvious approach is to translate this in the following manner: SELECT DISTINCT DO FROM (SELECT V01.DO DO FROM V01, V02 WHERE V01.DO V02.DO AND V01.D1 3 AND V02.D2 IN (SELECT V34.D3 D3 FROM V34 WHERE V34.D3 V02.D2 AND V34.D4 3) Notice the constraint v34.D3 v02.D2 which is instrumental for this to perform well on database systems that have poor optimizers.

Alternatively, this can be written by using a join approach: SELECT DISTINCT DO FROM (SELECT V01.DO DO FROM V01, V02, V34 WHERE V01.DO VO2.DO AND V01.D1 3 AND VO2.D2 V34.D3 AND V34.D4 3); This implementation should be in a form that is easier for most SQL optimizers, with regard to choosing a right execution plan. As seen later, one may rewrite any SDL expression in such a manner that one can apply this join approach instead of the above approach.

Before leaving the IN statements, it is illustrative to see how binding variables can be used inside nested SDL sets: {dO I [dl 3; $x dl] AND d2 IN d3 d4 $x can be translated as: SELECT DISTINCT DO FROM (SELECT V01.DO DO FROM V01, V02, V34 WHERE V01.DO V02.DO AND V01.D1 3 AND V02.D2 V34.D3 AND V34.D4 V01.D1); Finally consider: {dO I [dl 3 AND d2 IN d3 d4 $x $x dl] which is translated as: SELECT DISTINCT DO FROM (SELECT V012.DO DO FROM V012, V34 WHERE V012.D1 3 AND V012.D2 V34.D3 AND V34.D4 V012.D1); The above SDL query might have been more readable using alternative binding declaration: {dO I [dl 3 AND $x:=dl AND d2 IN d3 d4 $x which can be translated in the same manner as above.

-36- Consider the case where nesting is used inside a nested set: dO dl 3 AND d2 IN d3 d4 4 AND d5 IN d6 d7 The easiest and most straightforward approach would be to use SQL structure with IN to translate this and use multiple nesting in SQL as well. However, this SDL expression can also be solved with the join approach if the IN term is treated in a similar way as AND is treated: SELECT DISTINCT DO FROM (SELECT V01.DO DO FROM V01, V02, V34, V35, V67 WHERE V01.DO V02.DO AND V01.D1 3 AND V02.D2 V34.D3 AND V34.D4 4 AND V34.D3 V35.D3 AND V35.D5 V67.D6 AND V67.D7 One way to understand the SQL translation better is the following SDL equality: dO I dl AND d2 dO I dl AND dO IN dO I d2 Consider an example where there is a relation that associates an individual with its parents, e.g. pid, father, mother pid I diag "stro" AND father IN pid I diag "stro" Ex. al This query finds all individuals that have been diagnosed with stroke that in addition have their fathers as members in the set of individuals that have stroke. The father attribute, which is associated with an individual, is indeed a reference (RD) or a pointer to another individual. The domain of the father RD will imply the domain of the nested set, i.e. the default output dimension. Thus, to make the syntax as concise as possible, the invention SDL system also allows this same expression to be written as: diag "stro" AND father IN diag "stro" Ex. a2 The fact that the domain of each RD implies the PD is also used in the SDL GUI to support what applicants refer to as nested dimension drilling. Because of the -37naming convention of dimensions in SDL, it is natural to organize and present them visually in a hierarchical manner, i.e. all the dimensions that are related to a given PD are presented as leaves. Furthermore, a user can continue drilling from a RD into all the dimensions associated with the corresponding PD, select it and use drag-and-drop to compose nested queries in an easy manner. This provides similar experience to SDL query designers as for software developers that use OO-IDE tools to browse classes and their methods.

Now consider a more involved patient definition where in addition to having a relation that associates an individual with his parents, one has a relation that associates each individual with his children pid, child This VR can indeed be defined from the same source tables as the parent relation, pid, child father, pid u mother, pid With these relations in place, one can write: pid diag "stro.isch" AND father IN {diag "stro.isch" AND mother IN NOT diag "stro" AND child IN sex "male" AND diag "stro.isch" Ex. a3 This expression finds all patients that have been diagnosed with ischemic stroke that in addition have fathers and one or more male children that have been diagnosed similarly, but also, have mothers who have never been diagnosed with stroke. Similar queries might be of interest in the study of paternally inherited diseases. Again, it is illustrative to contrast SDL with the corresponding SSDL statement: -38- SELECT DISTINCT d.<pid> FROM [pid, diag] AS d, [pid, father, mother] AS p, [pid, child] AS c WHERE d.<pid> p.<pid> AND d.<pid> c.<pid> AND p.<father> IN (SELECT <pid> FROM [pid, diag] WHERE <diag> "stro.isch") AND p.<mother> IN (SELECT <pid> FROM [pid] EXCEPT (SELECT <pid> FROM [pid, diag] WHERE <diag> "stro.isch")) AND c.<child> IN (SELECT <pid> FROM [pid, diag] WHERE <diag> "stro.isch"); Ex. a4 Clearly, the succinct SDL notation of the present invention is much more intuitive and easily understood even though the operator is not expanded in the SQL code.

Nested queries are also very useful to form expressions with attributes from multiple types of objects, e.g. individuals and tissue samples. As an example of such, consider finding all male individuals that have DNA samples: pid I sex "male" AND sample IN sid type "DNA" AND location "roomA.freezer2" Ex. In the above statement, the sample attribute (RD) that is associated with each individual in [pid, sample], references a sample identifier, sid. The samples can of course be classified with whatever attributes that are available on them in relations such as the sample type and information on storage location, as shown here.

As a final example, consider nested sets as qualifiers in multi-attribute relationship, i.e. a relation. Here applicants draw an example from the genomics field.

Imagine that one wants to find all genes that are on the X chromosome that have protein sequence similarity above a certain threshold with any of the few hundred genes classified as gene protein coupled receptor (GPCR): gene chrom AND protsim.score 0.95 AND protsim.gene IN gene ontology "GPCR" Ex. a6 -39- For this query to be possible, at least three relations must exist, one storing data on gene locations, another relation storing gene ontology information (possibly with multiple classifications per gene) and another relation which stores the results of protein sequence similarity between all of the genes in the genome. Typically, a minimum threshold needs to be set on the similarity score to avoid storing an entry for every combination of gene pairs (40k x 40k rows). Such similarity score could for instance be generated by applying the Blast algorithm (Altschul, S. F. et al., "Issues in searching molecular sequence databases," Nature Genetics 6(2):119-129, 1994) on the protein sequences of the genes. Similarly, one could use binding variables to find genes with higher protein similarity to one class than another class of genes.

The SIZE set function Related to the IN statements and binding variables is the SIZE function on SDL sets. The SIZE function is simply to evaluate the size of a set and should mostly be useful in conjunction with the use of binding variables. Consider the following SDL statement: dO dl>O AND $x:=d2 AND SIZE{ d3 I d4 $x 2 The translation of an SDL statement with a SIZE function is straightforward: SELECT DISTINCT DO FROM (SELECT V01.DO DO FROM V01, V02 WHERE V01.DO V02.DO AND V01.DO 0 AND (SELECT COUNT(*) FROM (SELECT DISTINCT V34.D3 FROM V34 WHERE V34.D4 V02.D2)) 2 Notice the equivalence between the two SDL statements: dO dl AND $x:=d2 AND SIZE{ d3 d4 AND d3 $x 0 dO dl AND d2 IN d3 d4 The ALL keyword on records The SDL syntax has an inherent ANY or EXIST notation with regard to the expressions. Sometimes it may be desired to require that all of the tuples/rows in a relation fulfill a certain criteria. For this purpose one can use the ALL keyword in front of a record-operator which indeed is a shorthand for the following NOT syntax: dO I dl AND ALL[ d2>x dO dl AND AND NOT[NOT(d2>x)]) The ALL keyword simply provides a more readable syntax. Notice that NOT ALL should then be translated as (see rewrite rules in following sections): dO dl AND NOT ALL[ d2>x] dO dl AND NOT AND NOT[NOT(d2>x)]) dO dl AND (NOT[d2>x] OR dO dl AND NOT[d2>x] OR dl AND [NOT(d2>x)] Here a query -rewrite rule (see following section) has been used to eliminate double negation and negation on brackets.

General SDL expression Based on the short expressions above one may deduce a general approach for translating an SDL statement of the invention into an SQL statement. Consider now a general disjunctive expression: dO dl pi OR d2 4 AND NOT d3 5 dO SDLexprl OR SDLexpr2 The invention system translates this in the manner it translates the simple OR expression, e.g. by using the union approach. By using SDL2SQL to denote a function that translates SDL expressions to corresponding SQL statements, one obtains: SELECT DISTINCT DO FROM (SDL2SQL(SDLEXPR1) UNION ALL SDL2SQL(SDLEXPR2)) SELECT DISTINCT DO FROM SELECT V01.DO DO FROM V01 WHERE V01.D1 PI UNION ALL SELECT V02.DO DO FROM VO2 WHERE V02.D2 4 MINUS SELECT V03.DO DO FROM VO3 WHERE VO3.D3 5) Notice how the brackets enclosing the SQL code for SDLexpr2 are necessary because the precedence on UNION and MINUS in SQL is not the same as on NOT and OR in SDL.

-41 Now consider a general conjunctive SDL expression: dO dl pi AND d2 4 OR d3 5 dO SDLexprl AND SDLexpr2 Initially the invention system might translate this in the same manner as it did with the simple AND expression above: SELECT DISTINCT DO FROM (SELECT VA.DO DO FROM V01 VA, SDL2SQL(SDLEXPR2) VB WHERE VA.DO VB.DO AND VA.D1

PI);

To clarify this, expand the SQL statement: SELECT DISTINCT DO FROM (SELECT VA.DO DO FROM V01 VA, (SELECT V02.DO DO FROM VO2 WHERE V02.D2 4 UNION ALL (SELECT V03.DO DO FROM VO3 WHERE VO3.D3 VB WHERE VA.DO VB.DO AND VA.D1 PI); This approach works even when SDLexpr2 contains binding variables defined in SDLexprl although the two expressions cannot then be translated separately.

Figure 9 summarizes the general approach for translating disjunctive 900 and conjunctive 901 SDL expressions.

An alternative approach is to use nested SQL with IN statement, the same way as the invention handles negation. The IN translation approach is considered in the following example where binding variables are also used: dO [dl pi; $x d2] AND ([d3 5 AND d2 OR d3 do [dl pi; $x d2] AND SDLexpr2 With the IN translation approach this becomes: SELECT DISTINCT DO FROM (SELECT VA.DO DO FROM V012 WHERE VA.D1 PI AND V012.DO IN (SELECT V023.DO DO FROM V023 WHERE V023.DO V12.DO AND V023.D3 5 AND V023.D2 V012.D2 UNION ALL (SELECT V03.DO DO FROM V03 WHERE V03.DO V012.DO AND V03.D3 6 The drawback with both of these approaches becomes clear if one considers an extreme case where d3 6 is highly selective, much more so than dl pi. Because of -42the union operator and the binding variable, the cursors on v023 and v03 will be nested under the cursor on v012. Therefore, for each value of the v012 cursor, a lookup will be done on both v023 and v03. This lookup is approximately twice as expensive as a lookup only on v023. This example is further discussed later in the description of the OR distribution rewrite approach.

Query rewrite approaches In the previous section, Applicants showed how one could map most of the basic structures in the invention SDL syntax to corresponding SQL statements. On purpose, the discussion omitted the handling of brackets since brackets are indeed only to control the evaluation order. In this section, Applicants show rewrite rules for the parser 11 (Fig. which, amongst other tasks, eliminates all brackets and puts a subject SDL statement into conjunctive normal form (CNF) that is easily translated into an efficient corresponding SQL statement.

The rewrite rules of the invention are based on the following rules of algebraic logic: I. e e II. el AND e2 e2 AND el III. (el AND e2) AND e3 el AND (e2 AND e3) IV. el OR e2 e2 OR el V. (el OR e2) OR e3 el OR (e2 OR e3) VI. (el OR e2) AND e3 (el AND e3) OR (e2 AND e3) VII. (el AND e2) OR e3 (el OR e3) AND (e2 OR e3) VIII. NOT(el) OR NOT(e2) NOT(el AND e2) IX. NOT(el OR e2) NOT(el) AND NOT(e2) X. NOT NOT e NOT(NOT e) e Rules II and IV are based on the commutative law, rules III and V on the associative law and rule VI on the distributive law. Rule VII is not applied in the preferred embodiment of the present invention, however, it is listed for the sake of completeness. Rules VII and VIII are based on deMorgans's law. In particular, Applicants refer to rule VI as OR distribution.

-43- OR distribution O Consider the general SDL expression that is made up from three conjuncts: dO SDLexprl AND SDLexpr2 OR SDLexpr3 5 dO (SDLexprl AND SDLexpr2) OR (SDLexprl AND SDLexpr3) dO SDLexprl AND SDLexpr2 OR SDLexprl AND SDLexpr3 r-- 0Here the invention drops the brackets and insists that AND has higher precedence than C' OR in SDL as is the case in most computer languages. As an example consider: dO dl pi AND [d2 4 AND d3 0] OR d4 4 dO dl pi AND [d2 4 AND d3 0] OR dl pi AND d4 =4} According to the previous sections, using joins for AND and unions for OR, this example is translated into the following SQL statement: SELECT DISTINCT DO (SELECT V01.DO DO FROM V01, VO23 WHERE V01.DO VO23.DO AND V01.D1 PI AND V023.D2 4 AND VO23.D3 0 UNION ALL (SELECT V01.DO DO FROM V01, VO4 WHERE V01.DO V04.DO AND V01.D1 PI AND VO4.D4 4) Continuous application of the OR distribution results in an SDL expression in a conjunctive normal form (CNF). As an example, consider: dO dl AND (d2 OR d3 AND d4 OR d5 dO dl AND (d2 OR (d3 AND d4) OR (d3 AND d5) dO dl AND d2 OR dl AND d3 AND d4 OR dl AND d3 AND This is now easily translated to a joins for the conjuncts and unions for the disjuncts.

Consider the example from before: dO [dl pi; $x d2] AND ([d3 5 AND d2 OR d3 dO ([dl pi; $x d2] AND [d3 5 AND d2 OR ([dl pi] AND d3 6) -44- Based on the invention's standard translation approaches for AND and OR, one writes this as: SELECT DISTINCT DO FROM (SELECT V012.DO DO FROM V012, V023 WHERE V012.D1 PI AND V012.DO V023.DO AND V023.D2 V012.D2 UNION ALL SELECT V01.DO DO FROM V01, V03 WHERE V01.DO V03.DO AND V01.D1 PI AND V03.D3 6) Written in this way, an SQL optimizer 13 can easily choose an independent execution path for each of the two SQL parts, separated by the union operation, thus resulting in almost twice as low execution cost if d3 6 is highly selective as compared to the other criteria. The difference can be even more dramatic in scenarios where for instance d3 5 is also much more selective than dl pi, because with this SQL structure, the SQL optimizer 13 can choose to make the cursor on vO1 to be the most nested in both parts of the SQL statement, whereas, most SQL optimizers will not do that in the two previous SQL statements (see R. Ramakrishnan and J. Gehrke, "Database Management Systems", 2nd ed., McGraw Hill, (2000)).

Finally, it is interesting to see how OR distribution can help in expressions with negation: dO dl 1 AND NOT d2 2 OR d3 3 By using the IN translation approach this would map to: SELECT DISTINCT DO (SELECT V01.DO DO FROM V01 WHERE V01.D1 1 AND V01.DO IN (SELECT VO.DO DO FROM VO WHERE VO.DO NULL MINUS SELECT V02.DO DO FROM V02 WHERE V02.D2 2 UNION ALL (SELECT V03.DO DO FROM V03 WHERE V03.D3 The problem with this statement is that the part: "SELECT VO. DO DO FROM VO WHERE VO. DO NULL" can be quite costly since vO can be quite large (see the discussion on inclusion criteria for dimension). By using OR distribution the SDL expression becomes: dO dl 1 AND NOT d2 2 OR dl 1 AND d3 3 and the corresponding SQL translation results: SELECT DISTINCT DO (SELECT V01.DO DO FROM V01 WHERE V01.D1 1 MINUS SELECT V02.DO DO FROM V02 WHERE V02.D2 2 UNION ALL (SELECT V01.DO DO FROM V01, V03 WHERE V01.DO VO3.DO AND V01.D1 1 AND V03.D3 3) OR-distribution in record-operators The use of OR inside a record-operator can be a shorthand for defining two separate record-expressions or a way for setting two different conditions on the same dimension. In order to understand this consider the following cases: dO [dl 4 AND (d2 2 OR d3 dO [dl 4 AND d2 2] OR [dl 4 AND d3 In this case Applicants apply the OR-separation and translate it to SQL in the following way: SELECT DISTINCT DO FROM SDL2SQL( dO I [dl 4 AND d2 2] UNION ALL SDL2SQL( dO [dl 4 AND d3 2] Each of the SDL2SQL blocks is translated to an efficient join. In the next example the situation is different: dO [dl 4 AND (d2 2 OR d2 dO [dl 4 AND d2 2] OR [dl 4 AND d2 In this example one could use the rewrite as shown, however, this will result in a UNION ALL operation in the SQL statement. If the SQL optimizer 13 chooses to use dl as the outer-relation in the joins, the cost of evaluating this using this approach is two times higher than if the statement is translated directly to SQL, without the rewrite: SELECT DISTINCT DO FROM (SELECT V01.DO DO FROM V012 WHERE V012.D1 4 AND (V012.D2 2 OR V012.D2 If the SQL optimizer 13 chooses d2 as the outer-relation, there is however no difference, since index-lookup on a condition with OR is implemented using UNION, unless the database uses a full-scan on the table. The present invention however does -46not apply the rewrite since the SQL optimizer 13 can then choose to do so based on its cost estimates. Now, consider the following case: dO [dl 4 AND (d2 2 OR d2 dl dO [dl 4 AND d2 2] OR [dl 4 AND d2 dl Here the present invention does not apply this rewrite since both of the recordexpressions contain the same set of dimensions.

On the other hand consider: dO [dl 4 AND (d2 2 OR d2 d3 dO [dl 4 AND d2 2] OR [dl 4 AND d2 d3 Here the two record-expressions that result after the rewrite have different sets of dimensions, hence, it is necessary to apply the rewrite because the following straight forward mapping based on the first SDL form would be incorrect: SELECT DISTINCT DO FROM (SELECT V0123.DO DO FROM V0123 WHERE V0123.D1 4 AND (V0123.D2 2 OR V0123.D2 V0123.D3 The reason why this is incorrect is that the view v0123 assumes the existence of d3 even though d3 is not required. Translation of the rewritten SDL expression gives on the other hand: SELECT DISTINCT DO FROM (SELECT V012.DO DO FROM V012 WHERE V012.D1 4 AND V012.D2 2 UNION ALL SELECT V0123.DO DO FROM V0123 WHERE V0123.D1 4 AND V0123.D2 V0123.D3 To summarize, the invention applies the OR-distribution on record-expressions if it results in record-expressions with different sets of dimensions, i.e. if the conjuncts inside a subject record-expression have different sets of dimensions, the record-expression is separated into multiple record-expressions.

OR distribution and nested sets Consider now an SDL expression with an IN statement and a disjunctive expression in a nested set: -47- {dO I dl IN d2 I d3 3 OR d3 4 This can be translated in the following manner: SELECT DISTINCT DO FROM (SELECT V01.DO DO FROM V01, V23 WHERE V01.D1 V23.D2 AND V23.D3 3 OR V23.D3 4); If one changes this slightly by replacing the latter d3 by d4, one will have to apply the following rewrite in order to be able to use the same SQL structure: {dO dl IN d2 d3 3 OR d4 4 {dO dl IN d2 d3 3 OR dl IN d2 d4 4 This can be translated in the following manner: SELECT DISTINCT DO FROM (SELECT V01.DO DO FROM V01, V23

WHERE

V01.D1 V23.D2 AND V23.D3 3 UNION ALL (SELECT V01.DO DO FROM V01, V24 WHERE V01.D1 V23.D2 AND V24.D3 4) Hence, this shows that one can treat IN statements in the same way as record expressions, i.e. one can apply OR distribution within the nested SDL sets. Without this approach the expression shown above would have to be mapped using nested SQL or the following: SELECT DISTINCT DO FROM (SELECT V01.DO DO FROM V01 WHERE V01.D1 IN (SELECT V23.D2 D2 FROM V23 WHERE V01.D1 V23.D2 AND V23.D3 3 UNION ALL (SELECT V24.D2 D2 FROM V24 WHERE V01.D1 V24.D2 AND V24.D4 or alternatively as: SELECT DISTINCT DO FROM (SELECT V01.DO DO FROM V01,(SELECT V23.D2 D2 FROM V23 WHERE V23.D3 3 UNION ALL (SELECT V24.D2 D2 FROM V24 WHERE V24.D4 V2 WHERE V01.D1 V2.D2); As mentioned before, the drawback with these two approaches is that often the SQL optimizer 13 will not be able to handle this structure as well as the alternative -48proposed above. This approach is however useful when the nesting is preceded with negation, such as in the following example: {dO I [dl AND NOT d5 IN d2 d3 3 OR d4 4 OR merging Consider now a general structure of a CNF formatted SDL expression: dO I ANDexprl OR ANDexpr2 OR SDLexpr3 If any two conjuncts have the same set of dimensions, they can be merged together in the translation to SQL. A simple example explains this: {dO [dl 1 AND d2 2] OR [dl 3 AND d2 4] which can be translated in the following manner: SELECT DISTINCT DO FROM (SELECT V012.DO DO FROM V012

WHERE

(V012.D1 1 AND V012.D2 2) OR (V012.D1 3 AND V012.D2 This can give a factor of two in speed increase in cases where the SQL optimizer chooses full scan. Note however that the merge cannot be applied for conjuncts that differ by a NOT connective or by reference to different binding variables.

Negations Query rewrites related to negation can be done according to rules VIII X as well as rule II above. As an example consider the following SDL expression: {dO I NOT dl 1 AND d2 2 {dO d2 2 AND NOT dl 1 In the initial form the SDL expression is not easily translated to SQL using a left to right join approach, however, after the rewrite based on the commutative law, it is easily translated to: SELECT DISTINCT DO FROM (SELECT VO2.DO DO FROM V02 WHERE VO2.D2 2 MINUS SELECT V01.DO DO FROM V01 WHERE V01.D1 1); -49or to a corresponding SQL with IN keyword: SELECT DISTINCT DO FROM (SELECT V02.DO DO FROM VO2 WHERE V02.D2 2 AND V02.DO NOT IN (SELECT V01.DO DO FROM V01 WHERE V01.DO V02.DO AND V01.D1 1) The latter structure works also when the second part of the conjunction contains a reference to a binding variable defined in the other half, as mentioned earlier. The present invention therefore pushes negations to the end of any conjunct.

For instance if there is more than one negation: {dO I dl 1 AND NOT d2 2 AND NOT d3 3 This can be translated as: SELECT DISTINCT DO FROM (SELECT V01.DO DO FROM V01 WHERE V01.D1 1 AND V01.DO NOT IN (SELECT V02.DO DO FROM VO2 WHERE V01.DO V02.DO AND V02.D2 2) AND V01.DO NOT IN (SELECT V03.DO DO FROM V03 WHERE V01.DO V02.DO AND V03.D3 3) or into an SQL form using two MINUS operations.

If negation is in front of two or more disjunctive SDL expressions, the invention applies rewrite according to rule VIII, e.g.: {dO I NOT(el) OR NOT(e2) OR e3 dO NOT(el AND e2) OR e3 In this example, one saves one set subtraction from the "universe" as explained in previous sections. If negation encloses brackets, the present invention only applies rewrite if it contains a conjunctive expression according to rule IX. Before concluding discussion on rewrite approaches for negation, Applicants take one detailed example: {dO [dl;$x=dl] AND NOT(d2>$x OR d3 OR NOT(d4)) {dO I [dl;$x=dl] AND NOT(d2>$x) AND NOT(d3) AND d4 {dO d4 AND [dl;$x=dl] AND NOT(d2>$x) AND NOT(d3) Notice how the negation with the referral to the binding variable precedes the other negation because it has to be implemented with an IN structure whereas the latter term can be enforced by subtraction using MINUS: SELECT DISTINCT DO FROM (SELECT V04.DO DO FROM V04, V01 WHERE V04.DO NULL AND V01.DO NULL AND V04.DO V01.DO AND V01.DO NOT IN (SELECT V02.DO FROM V02 DO WHERE V01.DO V02.DO AND V02.D2 V01.D1) MINUS SELECT V03.DO DO FROM V03 WHERE V03.DO NULL Undefined binding variables The rewrite approaches suggested above, i.e. to translate the SDL expression into CNF, have an interesting side effect when it comes to the binding variables. Here Applicants explain how the invention treats binding variables that appear in a different conjunct than where they are declared. Consider for instance the following example: {dO OR AND ([d4 AND d3>$x] OR [d4 AND d3>$y] If one rewrites this SDL expression after performing OR distribution, one obtains: {d0 AND [d4 AND OR AND [d4 AND OR AND [d4 AND OR AND [d4 AND Notice that the SDL expression now consists of four conjuncts and that in the second and the fourth conjuncts, there are binding variables that are not defined within the same conjunct. One might think that the invention approach for implementing binding variables no longer holds, i.e. with join operation, since the scope of variables does not extend over the UNION statement in SQL that is used for implementing OR.

Applicants recognize, however, that conjuncts with undefined binding variables are FALSE, that is they can be eliminated from the disjunctive expression. The results therefore are: {dO [dl;$x=d3] AND [d4 AND d3>$x] OR [d2;$y=d3] AND [d4 AND d3>$y] Applicants also recognize that this expression could have been rewritten in another way: {dO I [dl OR d2;$x=d3] AND [d4 AND d3>$x] -51 This is because the SDL syntax allows disjuncts that arise because of different dimensions inside the SDL record operator. Consider another example: {dO OR d3) AND OR d5) This is a perfectly legal SDL expression, assuming that the dimensions d2 and d4 are of the same domain. Without rewrite approaches, it is non trivial to transform this to SQL since the relation (d0,dl,d2) is different from (d0,d3). One might try the following SQL code: SELECT DISTINCT DO FROM SELECT VA.DO DO FROM (SELECT V012.DO DO FROM V012 WHERE V012.DO NULL UNION ALL (SELECT V03.DO DO FROM VO3 WHERE VO3.DO NULL)) VA, (SELECT VO4.DO DO FROM VO4 WHERE V04.D4 V012.D2 UNION ALL (SELECT V05.DO DO FROM WHERE V05.DO NULL)) VB WHERE VA.DO VB.DO) Notice however, that referral to V012.D2 (boldface in the SQL statement) is illegal since it is not within scope there. Also notice that it would be impossible to refer to it as VA.D2 because the relation V03 does not have the D2 dimension and therefore, the relation VA only contains the DO dimension. The reader can verify the other approach that was described above for translating a general SDL expression to SQL, i.e. using the IN keyword, is also non trivial in this example (fake columns have to be used). The solution is however to apply OR distribution to the SDL expression before translation to SQL: {dO OR d3) AND OR d5) {dO [dl;$x=d2] AND OR [dl;$x=d2] AND d5 OR d3 AND OR d3 AND d5 Applicants recognize one conjunct term in the disjunction with an "undefined" binding variable and eliminate that term as discussed above. Hence one obtains: (dO I [dl;$x=d2] AND OR [dl;$x=d2] AND d5 OR d3 AND d5 This is easily translated to SQL with three joins that are unioned together.

Figure 10 illustrates the foregoing optimization and translation routines and procedures for optimized translation of SDL to SQL that uses a parser 11 in initial -52steps. This optimized translation has the benefit of generating SQL that is faster, especially on RDBMS where the SQL optimizer is not too good.

Aggregate functions The SDL language supports aggregate functions in a similar manner as in the SQL language. In the spirit of sparse syntax, the invention defines a way for using aggregates without any additional HAVING keyword as in SQL and the default grouping (GROUP BY in SQL) is defined implicitly, based on the output-dimension of the set. It is best to take a simple example: {dO I MAX(dl) x {dO I dl x This definition defines all dO for which the largest dl in relations with dO and d I is larger than x. Note that due to the "exist" nature of the SDL language, the above query is equivalent to the other form shown to the right. If another aggregate operator would be used, this is not the case. For instance the following SDL definition: {dO I MIN(dl) x {dO Idl x The former definition defines all dO for which the smallest dl in relations with dO and dl is larger than x, which of course is not equal to the latter definition that only requires some dl to be larger than x. Now let's see how the former definition can be translated into a SQL code: SELECT DISTINCT DO FROM SELECT V01.DO DO FROM V01 GROUP BY V01.DO HAVING MIN(VO1.Dl)>x Notice how the implicit grouping on the output-dimension in the SDL statement becomes explicit in the SQL code. Also notice how the aggregate constraint becomes a HAVING filter on the SQL relation V01. Let's take a more complicated example: {dO I AVG(dl) x WHERE d3 y GRBY d2 Here there are two types of constraints, a regular relational-expression (following the WHERE keyword) and an aggregate relational-expression.

Furthermore, there is additional grouping specification, i.e. the average is not only -53 calculated as a function of dO but the combination of dO and d2. The SQL translation Scould look like this: 0 SELECT DISTINCT DO FROM SELECT V0123.DO DO FROM V0123, V0123.D2 WHERE V0123.D3 5 y GROUP BY V0123.DO, V0123.D2 HAVING AVG(V0123.Dl)>x Note that the aggregate constraints show up in a HAVING clause whereas the CN regular constraints come in a regular WHERE clause. Also, notice the specification Son the grouping clause, i.e. both the implicit and the explicit dimension in the SDL CN statement. The use of the WHERE keyword is simply for making the syntax more readable and also to give instructions on what kind of constraint it is. Alternatively, the SDL syntax also supports an opposite setup, i.e. with a HAVING clause, e.g.: {dO d3 y HAVING AVG(dl) x GRBY d2 This SDL statement is fully equal to the previous SDL statement. However, the following statement is not: {dO AVG(dl) x AND d3 y GRBY d2 There are a few things worth mentioning here. First, Applicants have omitted the HAVING keyword since it is optional when the constraint starts with a calculated expression using an aggregate operator. Furthermore, the second constraint is also a HAVING constraint, as seen in the translated SQL below, and due to that fact, the dimension d3 will also appear in the GROUP BY statement (it is therefore optional to specify it in SDL): SELECT DISTINCT DO FROM SELECT V0123.DO DO FROM V0123 GROUP BY V0123.DO, V0123.D2, V0123.D3 HAVING V0123.D3 y AND AVG(V0123.D1)>x Care must be taken when translating the SDL statement when the conjunct has two terms, when a standard join approach is used (as compared to intersect approach).

Consider the following: {dO COUNT(dl) x AND d2 y By using nested table expressions the SQL translation could look like this: -54- SELECT DISTINCT DO FROM (SELECT VA01.DO FROM (SELECT V01.DO FROM V01 GROUP BY V01.DO HAVING COUNT(V01.D1) x VA01, V02 WHERE VA01.DO V02.DO AND VO2.D2 y); Notice how the aggregate grouping is applied before the join, because of a possible tuple-multiplication from the join process.

Aggregates can also be useful in conjunction with nested sets. Consider the following example: {dO COUNT(dO) x WHERE dl IN d2 d3 y Because of the COUNT aggregate and possible multiplication of tuples, as mentioned above, the join approach proposed for the nested set constraint does not work! Consider the following SQL code: SELECT DISTINCT DO FROM (SELECT V012.DO FROM V012, V34 WHERE V01.D1 V23.D2 AND V23.D3 y GROUP BY V01.DO HAVING COUNT(V01.DO) x The code above is an incorrect translation! For a record with an aggregate condition as well as a condition based on nested sets, one must use the alternative IN translation approach, e.g.: SELECT DISTINCT DO FROM (SELECT V01.DO FROM V01 WHERE V01.D1 IN (SELECT DISTINCT V23.D2 FROM V23 WHERE V23.D3 y) GROUP BY V01.DO HAVING COUNT(V01.DO) x Before concluding the discussion about aggregates, consider two more examples, i.e. binding variables defined from aggregates and a case were aggregates and dimensions are mixed in a calculated expression.

{dO [$x:=MAX(dl) AND $y:=MIN(dl)] AND [d2>$y AND d2<$x] In this example, the HAVING keyword is left out because it is optional. This can be translated as: SELECT DISTINCT DO FROM (SELECT VA01.DO FROM (SELECT V01.DO, MAX(VO1.D1) BX, MIN(VO1.D1) BY FROM V01 GROUP BY V01.DO) VA01, V02 WHERE VA01.DO V02.DO AND V02.D2>VA01.BY AND V02D2<VA01.BX) Notice how the binding variables are made part of the output from the nested table expression, i.e. the intermediate relation VA01. Finally, consider: {dO MAX(dl) d2 x {dO I HAVING MAX(dl) d2 x {dO I HAVING MAX(dl) d2 x GRBY d2] Notice the use of additional grouping caused by the use of d2 in combination with aggregate calculations. The SDL compiler automatically recognized this. This is one of the subtle points of aggregates, however, those who are well at home in SQL will understand this due to the nature of grouping and the calculation of aggregates (if grouping of d2 is not specified, there is no way to determine which d2 value to use, because there can be many tuples with the same value for dO but different value of d2). This becomes more clear when one looks at the translation: SELECT DISTINCT DO FROM (SELECT V012.DO FROM V012 GROUP BY V012.DO, V012.D2 HAVING MAX(V012.DO)+V012.D2 x); This example emphasizes the point that a calculated-expression with any aggregate must go into a HAVING clause. Of course, the same dimension could also appear in a WHERE filter and more grouping can be specified, e.g.: {dO MAX(dl) d2 x WHERE d2 0 GRBY d3] Figure 10 is a high level flow diagram of the SDL optimization procedure (by optimizer 13) according to the foregoing. Other similar procedures are suitable.

View union rewrite In the discussion on tables and metadata, Applicants demonstrated how the SDL system uses its metadata to construct virtual relations by combining all the tables that contain the proper dimensions with the union operator. That approach causes the union operator to be applied before the joins are calculated (unless the SQL optimizer does the same rewrite as Applicants are proposing here). Consider the following expression: dO dl AND d2 Without the loss of generality, Applicants assume that tables A and B store relations (d0,dl) and that tables C and D store relations (d0,d2). One can indeed introduce a -56- "special" dimension, T that denotes the table in which the tuples (records) are stored.

Then the expression above can be written as: dO [dl AND (T=A OR AND [d2 AND (T=C OR Notice that this specifies a constraint on the dimension T, a constraint that is equivalent to "no constraint" given the assumptions about the data stored in the system. With the invention's regular SQL translation approach, this would be written as: SELECT DISTINCT DO FROM SELECT V01.DO DO FROM V01, V02 WHERE V01.D1 NULL AND V02.D2 NULL AND V01.DO V02.DO); SELECT DISTINCT DO FROM SELECT V01.DO DO FROM (SELECT V01.DO DO FROM (SELECT DO, D1 FROM A UNION ALL SELECT DO, D1 FROM B) V01, (SELECT DO, D2 FROM C UNION ALL SELECT DO, D1 FROM D) V02 WHERE V01.D1 NULL AND V02.D2 NULL AND V01.DO V02.DO); Now consider if OR distribution is applied on the special table dimension, T: dO I [dl AND T=A] AND [d2 AND T=C] OR [dl AND T=A] AND [d2 AND T=D] OR [dl AND T=B] AND [d2 AND T=C] OR [dl AND T=B] AND [d2 AND T=D] The corresponding SQL code takes the following form: SELECT DISTINCT DO FROM SELECT V01.DO DO FROM (SELECT V01.DO DO FROM A) V01, (SELECT DO, D1 FROM C) VO2 WHERE V01.D1 NULL AND VO2.D2 NULL AND V01.DO V02.DO) UNION ALL SELECT V01.DO DO FROM (SELECT V01.DO DO FROM A) V01, (SELECT DO, D1 FROM D) V02 WHERE V01.D1 NULL AND VO2.D2 NULL AND V01.DO V02.DO) UNION ALL -57- SELECT V01.DO DO FROM (SELECT V01.DO DO FROM B) VO1, (SELECT DO, D1 FROM C) V02 WHERE V01.D1 NULL AND V02.D2 NULL AND V01.DO V02.DO) UNION ALL SELECT V01.DO DO FROM (SELECT V01.DO DO FROM B) VO1, (SELECT DO, D1 FROM D) V02 WHERE V01.D1 NULL AND V02.D2 NULL AND V01.DO V02.DO After the rewrite, each conjunct is implemented with a join of only two tables, i.e. the relation view v01 does not have to be implemented with the union operator.

Instead, the union is carried out because of the disjunctive nature of the whole expression. The union has therefore been moved further back into the evaluation, i.e.

after the join whereas it is before the join in the original implementation. Notice that this allows the optimizer 13 to choose a separate execution plan for each join, hence it is more likely to perform well if the tables have different sizes, cardinality and indices.

Finally, it is worth mentioning that view union rewrite does not require the definition of a special dimension for table names. Those skilled in the art should see that the bookkeeping for this distribution rewrite could be carried out where a conjunct is translated to SQL code.

Figure 11 is a flow diagram of a view-union rewrite optimization procedure (by optimizer 13) according to the foregoing. Other similar procedures are suitable.

Automatic record operator enforcement The SDL syntax uses square brackets to denote a record operator. The record operator is used to enforce more than one criterion on a single relation instance (tuple/record). There are cases where the nature of the data is such that two expressions, with and without a record operator, are equivalent. This is explored next along with a rewrite optimization strategy that utilizes this fact.

Applicants assume that one has some conjunctive expression: dO dl AND d2 AND d3 AND d4 One can group all dimensions in a conjunct together into a set, R, where the following holds: each dimension in R takes part in the same relations as any other -58dimension in R, and for all those relations, the output dimensions (dO) is declared with its multiplicity as "unique". As an example, ifdl, d3 and d4 only exist in a single relation (dO, dl, d3, d4) where dO is unique, then the expression above can be rewritten as: dO I d2 AND [dl AND d3 AND d4] This will change the SQL translation from a 4 table join into a 2 table join, which can be dramatically much faster. A practical example of this is where one has a singleton relation with information on individuals such as (pid, sex, dob, dod), i.e.

demographic information. Consider a typical query: pid dob 1950 AND sex=male AND diag:stroke AND datedob 30 This expression could be automatically rewritten to: pid I [dob 1950 AND sex=male] AND diag:stroke AND date-dob 30 Notice that since there can be multiple records in (pid, diag, date) for each patient, diag and date would not automatically be locked into a record. In this case, the user is however, most likely trying to make the following query: pid [dob 1950 AND sex=male; $x=dob] AND [diag:stroke AND date-$x>30 Notice that since the dimension for date of birth, dob, is not in the same relation as date, it cannot be placed into the record operator, except through a binding variable.

As another example take the following query: pid I dob 1950 AND sex=male AND dod-dob 90 Because all of the four dimensions that appear in this conjunct only exist in one and only one relation and where the pid is unique, this could be automatically rewritten as: pid [dob 1950 AND sex=male AND dod-dob 90] As for the view union rewrite, automatic record operator enforcement is easily implemented where a conjunct is translated to SQL. Each SQL cursor that is generated for the conjunct join can be inspected and all cursors that refer to the same table(s) can be merged if the output dimension of those cursors is the same and it is declared unique. In other words, this would eliminate unnecessary self-joins. The -59automatic record-operator enforcement is really an optimization based on semantic equivalence, i.e. the optimizer 13 can recognize unique constraints on the relations it is using and rewrite the query into a form that has faster implementation than the most natural translation of the query as expressed by the user.

Greedy record locking and structured dimensions Sometimes it may be desired to have the optimizer 13 setup in such a mode that it enforces record locking even though the user does not specify it. The scenarios where this may be desirable is when the user is assumed to be inexperienced (unfamiliar with the record-operator) and the data is in such a form that it is most natural to constrain it as records, i.e. constraints on separate dimensions in the same relation make most sense on the same tuple (row). Another scenario is where the data volumes are very large such that forgetting record locking can lead to prohibitively expensive queries. Here Applicants consider a feature in the SDL invention that aims to simplify the query and ensure better performance. Applicants refer to it as greedy record locking.

Consider the following example: gid I expr.tissue=:brain AND expr.level 0.9 The relation that stores these dimensions is (gid, expr.tissue, expr.level) and there is not a unique constraint on gid. Given this fact, it is not semantically correct to enforce record locking on the two dimension constraint in the set definition. However, due to the nature of the data in this example, and there are many equivalent scenarios, the user most likely wanted to query the following: gid I expr.tissue=:brain AND expr.level 0.9 If one labels these dimensions, expr.tissue and expr.level, in such a manner that defines them to belong to a structure, then the optimizer 13 can automatically enforce record locking. The fact that two dimensions are defined as a structure should also require that they do not exist in isolation, in relations where either is missing.

There are multiple ways to define structure groups for dimensions in the metadata, e.g.

by labeling each dimension with an identifier of a structure group that could be stored as an additional column in 401 (Fig. The rules for record-enforcement are defined Sin the present invention in the following manner as illustrated in Fig. 12.

O Dimensions that fall into the same conjunct (after OR-distribution) are collected at 1201. Dimensions that are not enclosed with a record-operator are labeled 5 and inspected with regard to merging them into other records that form the conjunct at step 1202. The structures that are of interest are those that are associated with dimension fingerprints of labeled records, yet dimension fingerprints that are unique 0in the list of labeled records, and are also associated with other different records in the CK1 conjunct (step 1203). Note that in this context, a record can refer to single dimensions or relational-expressions in a conjunct, e.g. thus dimensions that appear together in a calculated-expression (but not enclosed within a record-operator) are treated together, i.e. they have to belong to the same relation. If a labeled record (dimensions) can be merged with one and only one of the other records unambiguously), it is put into that record at step 1204.

Record-operator enforcement can be a) none, by ignoring structure information, i.e. apply only when semantically correct, b) regular, by applying as described above when structure info is found, or c) greedy, by assuming all dimensions that exist in the same relation to belong in a structure. Consider an example: gid I loc.start>1000 AND loc.stop<2000 AND loc.chrom=l Assume a relation of the form (gid,loc.start, loc.stop, loc.chrom). In this example, assume that a gene could have been mapped to multiple locations on the genome.

Therefore, automatic record-operator enforcement would not be possible. However, if these dimensions have been assigned to a structure then the optimizer 13 should be able to treat this as: gid I [loc.start>1000 AND loc.stop<2000 AND loc.chrom=l] which is probably what the user wanted. If however the user writes: gid I [loc.start>1000] AND [loc.stop<2000] AND loc.chrom=l -61 Here the optimizer 13 would not enforce any additional record-locking, because loc.chrom could not be unambiguously put into a record. Likewise: gid loc.start>1000 AND loc.stop<2000 AND loc.chrom=l AND loc.chrom=2 This is a silly example, however, it is used to emphasize that loc.chrom could not be locked with the other dimensions because there are two instances of it in the same conjunct.

Now consider the following: gid I [loc.start>1000 AND loc.stop<2000] AND loc.chrom=l AND name='A*' Assume that the dimension name is in a separate relation (gid, name). Because loc.chrom is in a structure with loc.start and loc.stop this will be rewritten by the optimizer 13 as: gid [loc.start>1000 AND loc.stop<2000 AND loc.chrom=l] AND name='A*' It should be noted, that there are other ways to define the rules for recordoperator enforcement. In this embodiment, the rule is in such a manner that nonrecord constraints are merged into other records if they fulfill certain rules.

Alternatively, the record enforcement could only be applied to constraints that are not enclosed by record operators. The former approach gives however a nice feature that will become clear when considering a general case: dO I exprl AND record-expr AND expr2 The expression inside the brackets could for instance be a user generated expression and the expression outside the brackets could be additional constraints generated automatically by an application. These automatically generated constraints could for instance be window time constraints or window spatial constraints. If the constraints in expr2 are stored in a relation with the record-expression, the rules of this embodiment allow additional machine generated constraints easily to be combined with any type of user-generated expressions that may contain records.

Finally, consider an example with nested sets: 0- -62-

O

S{ pid fid IN pid I $d:=dob AND mid IN pid dob>$d Typically, a genealogy relation should be a singleton (pid, fid, mid) and the optimizer should therefore be able to rewrite this no matter what. If the genealogy is, however, allowed to store multiple possible parent relationships then it would nevertheless make sense to define fid and mid as a structure. The optimizer 13 would be able to rewrite Sthe query that finds all individuals that have older fathers than mothers as: pid I [fid IN pid I $d:=dob AND mid IN pid NC dob>$d The point here is that the IN constraint is simply treated as a regular constraint. Of course, the same rewrite rules apply within each nested set.

Pivoted dimensions Next discussed is an extension to the standard SDL syntax (see the basic SDL syntax specification below) that allows the SDL language to treat dimensions in a hierarchical manner. This extension requires small changes to the meta data presented earlier and minor changes to the SDL "calc-expr" definition. As shown below, the benefit of this extension is to allow measurements to be classified in a hierarchical manner without using the record operator.

The extension redefines the calc-expr such that it includes "dimension:code".

To clarify this, Applicants take an example of an SDL expression that utilizes this extension: pid meas:bloodpressure.high 120 In this example, all measurements have been combined into what looks like a single dimension. Without this extension, the expression would have to be written as: pid I meas:bloodpressure.high AND value 120 The data could also have been formatted such that each type of measurement would reside in a separate dimension, e.g.: pid bloodpressure.high 120 With the invention extension, multiple dimensions are in effect pivoted into a single column and an additional column is needed to store its value. This can also be -63 referred to as vertical representation. Unless the values are stored as strings and converted to the proper domain data type on the fly, all the dimensions that are pivoted together should be of the same data type. To support this extension, it is necessary to augment the meta data. One way to do that is to store an additional column in the table domains, SQLpivoted_type, and an additional column in dim2rel, pivoted_value_column. The SDL system 100 would then recognize dimensions that have non null values in value_column and treat them appropriately when translating the SDL expression to SQL.

Applicants rewrite the first SDL expression in a more general manner: dO I dl:a.b 120 This expression would be translated to the following SQL code: SELECT DISTINCT DO FROM SELECT VO1X.D DO FROM V01X WHERE (VO1X.DI LIKE OR VO1X.D1 AND VO1X.X 120) Here the column name of the pivotedvaluecolumn is and the virtual view that includes dO, dl and x is denoted with vOlx. It should be noted that although this pivoting extension provides additional flexibility and extends the SDL schema such that it encompasses "all" relational schemas, e.g. horizontal, vertical, and partially pivoted data structure, it is usually more efficient to store dimensions in separate columns. However, by generalizing this approach, it is possible to hide the actual implementation of the relational schema and the SDL language itself simply works with logical or virtual relations.

For implementing access security for data and project separation, there are also implementation choices because of the SDL metadata abstraction. Access to data belonging to different projects can be controlled through the metadata manager. For instance, data from two different projects can be separated into two different SQL tables. It is trivial to support such logic by, for instance, adding access control data to table 402. An alternative approach would be to add an additional column to every data table in the system and let the SDL2SQL query generator include additional WHERE constraint on each cursor, in the spirit of the above code generation for pivoted dimensions. One way to view this is as if every logical relation has a a) C4 -64dimension specifying access level and every constraint on dimension is turned into a record expression that combines the original constraints with the system generated access constraints. This is similar to the approach taken in the so-called "Virtual Private Database" in Oracle.

The subject of adding additional dimensions to relations not only has relevance for implementing project access restrictions but also in controlling access and scope to federated data. Furthermore, this subject is related to record-operator enforcement, if one considers the additional scope dimensions as belonging to a structure with all other dimensions. However, ambiguity rules in record-enforcement have to be revised and these rules would also have to propagate into nested subqueries.

SDL syntax specification set-definition: setname sdl-set setname(parameters) sdl-set sdl-set: dimension I expression sdl-set sdl-set sdl-set sdl-set constantlist constantlist: constant constantlist, constant expression: code-expression expression AND expression expression OR expression NOT expression (expression) record-expression] record-expression; bindinglist] O aggregate-record-expression aggregate-record-expression GRBY dimensionlist ALL[ record-expression] 0 ALL[ aggregate-record-expression C1 ALL[ aggregate-record-expression GRBY dimensionlist SWITHIN( variablelist; constant) parameterlist: parameter parameterlist, parameter dimensionlist: dimension dimensionlist, dimension parameter: calc-expr bindinglist: variable dimension bindinglist, variable dimension variablelist: variable variablelist, variable record-expression: code-expression record-expression AND record-expression record-expression OR record-expression -66- NOT record-expression (record-expression) aggregate-record-expression HAVING record-expression HAVING record-expression WHERE record-expression record-expression HAVING record-expression record-expression WHERE record-expression code-expression: dimension=:code value dimension:code value relational-expression setname(parameterlist) dimension IN sdl-set variable calc-expr relational-expression: calc-expr rel-op calc-expr rel-op: calc-expr: constant domain code value) variable parameter dimension (calc-expr) calc-expr calc-expr calc-op calc-expr FUNCTION( calc-expr) -67aggregate-op(dimension) SIZE sdl-set calc-op: aggregate: id-code-aggregate set-code-aggregate id-code-aggregate aggregate-op dimension aggregate-op dimension; record-expression] set-code-aggregate set-aggregate-op dimension; model-expression set-aggregate-op dimension; model-expression; record-expression aggregate-op: COUNT, DISTINCT, AVG, STD, VAR, MAX, MIN, MEDIAN, FIRST, LAST It should be emphasized that this syntax is simply provided to explain the fundamentals of the SDL language and facilitate the creation of a basic SDL system of the present invention. This basic version of the SDL syntax does not need to entail the SDL invention and does for instance not cover all aspects of binding variable syntax, macros, SQL object-relational extensions nor integration of SDL with SQL through "syntactic gates", as described in following sections.

Bidirectional language inlining Because the SDL language maps to SQL, by definition SQL has all the expression power of SDL. Indeed, SQL has much more expressive power than SDL, however, SDL has much simpler syntax and therefore provides more readable and shorter expressions in the cases where SDL is suitable (the set-definition use-case).

The expressive power of SQL and its wide availability makes it very desirable to -68merge the two languages. In the following discussion the combined language of SQL and SDL will be referred to as SSDL.

In order to merge SQL and SDL two things are necessary: First, a mechanism is needed to map between the meta-data of SDL and the meta-data of SQL (the table and column names) and secondly, the ability to intertwine the SDL and SQL syntax is needed. This invention describes a way to satisfy these needs and to accomplish this in a way that gives an easily readable syntax which is also easily parsed using a "passthrough" compilation technique, i.e. the compiler does not have to parse and compile the whole SQL syntax. This simplifies the construction of the SSDL compiler.

Integration of the languages in SSDL is in two ways, i.e. one can inline SDL from within a SQL code and also inline SQL from within SDL. This indeed also means that one can use SDL code from within SQL which then again uses SQL code and vice versa. This leads to a double-recursive implementation of the SSDL compiler. As before, examples are used to explain the implementation of the SSDL language.

Consider the following code: select from Tablel as MEAS where MEAS.value 3 and MEAS.id IN pid I [diag=:icd9.stroke AND $d:=date] AND [drug=:act.tPA AND date>$d]} and MEAS.date 2000 This query is a simple selection of measurements from a relational table, Table 1. The rows that are selected are constrained in two ways: first, their date has to be larger than the year 2000 (a regular SQL WHERE clause and secondly, the ids (identifiers) have to belong to the set defined with the SDL expression. Notice how the curly-brackets are used to define the start and the stop of the SDL code. Curlybrackets are not part of the SQL syntax, thus the SDL code can be identified and compiled as described in this invention using SDL2SQL, and substitute the corresponding SQL code into the SQL statement.

In the example above, there was no mixing of the metadata from the two languages. For such mixing of metadata, consider the following example: select MEAS(*), D.<dob> D.<dod> from Tablel as MEAS, [pid, dob, dod] as D where MEAS.value 3 and MEAS.id IN pid I [diag=:icd9.stroke AND $d:=date] AND -69- [drug=:act.tPA AND date>$d] and MEAS.date 2000 and D.<pid> MEAS.id Note a new bracket-operator, <dimension>, is used to map an SDL dimension to a SQL column name. Furthermore, reference to SDL virtual relations through the SDL record-operator (relation-operator) is allowed. An alternative syntax for [pid,dob,dod] could have been [pid AND dob AND dod]. In order to refer to a virtual-relation, a SQL binding variable (cursor) needs to be assigned to it, D in the example above.

Again, it is worth noting that the bracket-operator, introduced in this invention to map between the SDL and the SQL meta-data, is easily recognized by a preprocessor and distinguished from comparison in SQL expressions. This is most easily accomplished by insisting that a binding variable cursor precedes the bracketoperator, thereby also insisting that a dot precedes it, e.g. dim The other symbols that the preprocessor needs to recognize are the squarebrackets used to denote the virtual relations. Since square brackets are not part of the standard SQL language, it is very easy to identify them in the SQL statement with a preprocessor and replace them with a corresponding SQL code that defines the virtualrelation.

The bracket-operator, is really a mechanism to translate from SDL dimensions to SQL column names. SDL dimension names do not necessarily comply with naming conventions of SQL column names. For instance, dots are allowed in SDL dimension names, however, they are illegal in SQL column names. Also, because virtual relations of dimensions are generated at runtime from the dimensions fingerprint, a dimension may refer to multiple different tables and columns. There are many ways to implement a mechanism that issues legal SQL names that have a oneto-one mapping with SDL dimension names, e.g. by hashing, a lookup-table etc., and it should be unnecessary to elaborate on this mechanism in the teaching of this invention.

Alternatively, SQL can be used from within an SDL statement. Consider the following: dO I [select Tl.cl as MAX(Tl.c2) as from T1 group by cl] AND [dl Applicants refer to this approach of using SQL inside SDL as the virtualrelation approach, since, instead of letting the record-operator define a relation based on the dimension fingerprint (including the output-dimension), a SQL select statement is put into the record operator instead. Actually, the select statement is not really a pure SQL code, but an SSDL code since it also uses the bracket-operator to define column names, such that the SDL code generator can use the relation defined with the select statement. It is mandatory to have a column that is named after the outputdimension in order to generate the implicit join. The select statement can return more columns and they can be used within the SDL statement as binding variables as shown above. Again the SDL parser 11 can recognize the SQL (SSDL) code within the record operator because "SELECT" is not an allowed keyword in SDL and it is assumed that no dimension is allowed to have that name. The bracket-operator is easily detected in the select statement although the invention also allows the bracketoperator to be preceded with an "as" keyword, in addition to a dot as mention above.

Before leaving this example, it can be informative to see how the overall SQL statement from the SDL/SSDL compiler could look: SELECT DISTINCT DO FROM SELECT SQL1.DO FROM( select Tl.cl as DO, MAX(Tl.c2) as BX from T1 group by cl )SQL1, V01 WHERE SQL1.DO V01.DO AND V01.D1 SQL1.BX); Another way of referring to SQL within an SDL statement is the nested setapproach shown in the following example: pid diag=:stroke AND hospid IN {select cl from Tablel) In this example a nested set is defined fully by a SQL statement. For this to work, the data type of the table column cl must be compatible with the domain of the corresponding dimension, here hospid. Furthermore, the select statement can only return a single column.

Yet another way of using SQL within SDL is to use an SQL-value gate for calculations, e.g.: -71 pid [diag=:stroke AND date>SQLV(select MAX(LOG(cl)) from Ti)] This approach is useful when SQL is not used to provide additional attributes on the objects defined by the output-dimension, but rather for comparison with some attributes such as shown here. Needless to say, the SQLV( function and its string parameter that contains a SQL code are easily recognized by a preprocessor.

As mentioned earlier, SSDL supports bidirectional language inlining and thus the inlining can be on multiple levels of nesting, e.g.: select MEAS(*), D.<dob> D.<dod> from Tablel as MEAS, [pid, dob, dod] as D where MEAS.value 3 and MEAS.id IN pid I [diag=:icd9.stroke AND $d:=date] AND [drug=:act.tPA AND date>$d] AND [select T2.cl as <pid> from Table2 as T2]} and MEAS.date 2000 and D.<pid> MEAS.id In this example, an SDL statement is inlined in the SQL part of the SSDL code and within that SDL statement, there is a SQL code being used according to the virtualrelation method.

The present invention implements an SSDL compiler (Figs. 13 and 14) with what Applicants refer to as a pass-through approach, i.e. the compiler does not have to understand the SQL language. A preprocessor is used that identifies the symbols that indicate the switching between the languages, i.e. a start of inlining. As seen from the previous examples, these are: "as from within SQL (SSDL) and "[select SQL/SSDL]", "{select SQL/SSDL}", and "SQLV(SQL/SSDL)". The pass-through approach has the benefit of ignoring any parsing and compilation of the SQL in the SSDL statement. Not only does this simplify the development of the SSDL compiler, but it also makes it much more independent of the SQL flavor (version) being used. Alternatively, one can create a single parser that has a full syntax specification for both of the languages.

It is important to note that common table expressions (defined using the WITH keyword in SQL) and views can only appear in the beginning of an SQL statement.

Therefore, a pass-through compiler that possibly has to compile multiple blocks with -72- SDL and SSDL code needs to collect them and place them before the overall result statement. The overall statement will contain both "uncompiled SQL code" (the passthrough code) as well as SQL code generated from SDL code blocks.

Figure 13 and figure 14 summarize the previous discussion in diagrams that show a preferred embodiment of a high level logic for SDL/SSDL pass-through compiler that supports bidirectional inlining. Worth noting are the steps that collect the SQL code of the inlined statement and the corresponding common table expressions, e.g. 1301 and 1401. These two steps are the cause for double-recursion in the compiler.

Default dimension prefix In many scenarios it can be useful to view dimensions as attributes of complex objects. In such cases it can often be intuitive to arrange the dimensions into a folder hierarchy, such that all the attributes for a given object fall into the same folder.

Attributes of different objects would then have different parent folders. If the names of the dimensions reflect this folder hierarchy, then all the dimension attributes in the SDL expressions will repeat the hierarchy structure. This leads to unnecessary long syntax. Default dimension prefixing is a special namespace technique intended to shorten the SDL syntax and make it more readable at the same time.

Let's take an example of SDL expressions with and without using default dimension prefixing. Consider the following expression: Sobjl.id I objl.al x OR [objl.a2 y AND objl.a3 0] Here the dimension aO represents the ids (identifiers) of the objects. By using default dimension prefixing the above will look like this: objl::id I al x OR [a2 y AND a3 0] Thus, by using double colon "after the dimension prefix objl, the invention makes objl the default prefix of all the dimensions (attributes) that belong to relations with the output-dimension.

Consider now an example with a nested set: -73- S{ objl.id I objl.al x AND objl.a2 IN obj2.a0 obj2.al z} This can be rewritten using default dimension prefixing as: objl::id al x OR a2 IN obj2::a0 al z t 5 Notice that the clause al x refers to a different dimension than the clause al z, i.e.

they refer to objl.al and obj3.al respectively. Thus, the nested set has a different namespace than does the outer-set.

Sometimes, it may be desired to refer to dimensions outside the current C namespace. There are two mechanisms to provide such support. First, if the dimension name that is generated by applying the default prefix is not existent in the metadata, it could be looked up in the global namespace without the dimension prefix.

Alternatively, in the spirit of naming conventions used for file names in most file systems, it is possible to use backslash dot-notation to refer outside the default-scope, i.e. "\dim" or It should be noted that default dimension prefixing has minimal impact on the SDL compiler and it mostly impacts the metadata manager, i.e.

if the parser can provide the metadata manager the namespace, based on the prefix on the current output-dimension, the metadata manager can be modified to recognize the dimensions based on the rules specified above.

Default dimension prefixing can also be used to generate shorter syntax for virtual relations in SSDL. As an example, consider: select from [objl.id, objl.al] This can be made equivalent to: select from [objl::id, al] Other equivalent syntaxes are also possible such as: select from objl:: [id, al] or simply select from objl[id, all A corresponding notation could then apply for records within SDL expressions, e.g.: objl.id I objl.al x OR objl[a2 y AND a3 0] -74- It is emphasized that the exact syntax is not really the issue here but the concept of using namespaces within SDL and SSDL.

Template functions in SDL The fact that SSDL is a language that combines the metadata of SQL and SDL and has all the expressive power of SQL, gives a lot of flexibility in terms of developing advanced functions into the SDL language. Applicants will refer to these functions as template functions because, during SDL compilation, they are expanded into SSDL code before the final compilation of the SDL code. There is one thing that makes these SDL template functions unique and that is the notion of context, i.e. they can be defined in terms of parameters that are not really part of the input parameters or constraints. The best way to describe this is to take an example.

Consider the following SDL statement: gene::id I func=:gpcr AND OVERLAPS({snp::id I nonsilent))} This query is supposed to find all genes that overlap with SNPs that are of type nonsilent. The definition of the OVERLAP template function assumes certain naming conventions, i.e. that spatial positions of genes and SNPs are stored as (gene.id, gene.start, gene.stop) and (SNP.id, SNP.start, SNP.stop), respectively.

The compilation of the template function results in the following SDL statement: gene::id I func=:gpcr AND [select G.<gene.id> from [gene::id, start, stop] as G, [snp::id, start, stop] as S where S.<snp.id> IN snp::id nonsilent and SQLOVERLAPS(G.<gene.start>,G.<gene.stop>,S.<snp.start>,S.

<snp.stop>)=l] Here the OVERLAP statement has been replaced with an SSDL code using the virtual relation approach. Also notice the virtual relation labeled as G that refers to gene::id, although there is no mention of gene::id within the parameters in the OVERLAP SDL template function. This is what the Applicants refer to as the unique context feature of the template functions (somewhat similar to the "this" keyword in object-oriented languages). Another way to look at this is the "implicit constraint" nature of SDL, i.e.

gene.id could have been passed to the template function, however that would have resulted in longer syntax. Instead the invention allows the template function definition to refer to a context. There are many ways to define the terminology for this.

The present invention proposes the following: OVERLAP(SDL-set [select G.<OUTDIM(@this)> from [PREFOUTDIM(@this), start, stop] as G, [PREFOUTDIM(@1), start, stop] as S where S.<OUTDIM(@1)> IN @1 and SQLOVERLAPS(G.<PREFIX(@this).start>,G.<PREFIX(@this).stop >,S.<PREFIX(@1) .start>,S.<PREFIX(@1) .stop>)=l] Both @this and @1 refer to sets, the set in which the OVERLAP template is used and the set that is the input to the OVERLAP template function, respectively. OUTDIM), PREFIX(, and PREFOUTDIM( are macros that are easily understood by comparing the definition of OVERLAP with the code in the previous example. Note that the function SQLOVERLAPS needs to be defined in the RDBMS 102.

Take one more example of a template function that also detects an operator and a parameter: gene::id NUMOVLAPS({snp::id I nonsilent})>10 AND chrom=l This query is meant to find all genes that overlap with more than 10 nonsilent SNPs.

In the present invention, definition for this template function would be: NUMOVLAPS(SDL-set SDL-relop @2 SDL-calc-expr @3 [select G.<OUTDIM(@this)> from (select G.<OUTDIM(@this)> from [PREFOUTDIM(@this), start, stop] as G, [PREFOUTDIM(@1), start, stop] as S where S.<OUTDIM(@1)> IN @1 and SQLOVERLAPS(G.<PREFIX(@this).start>, G.<PREFIX(@this).stop>,S.<PREFIX(@1) .start>, S.<PREFIX(@1).stop>)=1) GROUP BY S.<OUTDIM(@1)> HAVING count(G.<OUTDIM(@this)>) @2 @3] The compilation of the template function results in the following SDL statement: -76gene:;id I [select G.<gene.id> from (select G.<gene.id> from [gene::id, start, stop] as G, [snp::id, start, stop] as S where S.<snp.id> IN snp::id I nonsilent and SQLOVERLAPS(G.<gene.start>, G.<gene.stop>, S.<snp.start>, S.<snp.stop>)=l) GROUP BY S.<snp.id> HAVING count(G.<gene.id>) 10] The relational operator, and the calculated expression, are shown in bold in the final SDL code, just for clarification. Notice that the same SQL function is being used as in the previous template function. As a final remark, the present invention describes only two basic types of template functions. The main novelty is the use of the context in the compilation of the statement and the use of SSDL to merge SDL with SQL code.

Primary dimension and default output-dimension Another technique for creating a shorter syntax, that is related to default dimension prefixing, is default output-dimension. It has been mentioned in the above that the SDL server or the SDL editor can be configured in such a manner that if output-dimension is omitted from the set declaration, the system 100 will use a predefined default output-dimension. This concept can be generalized such that it applies also to nested-sets and SDL sets defined in SSDL. Indeed, this matter touches a subtle point in the definition of the SDL language and the blurry difference between domains and dimensions (primary dimensions). As above, this is explained through examples. Reference is made to relationl, table 300 of Fig. 3, which defines demography and genealogy relationships for individuals (pids).

Consider now the definition of all male individuals that have long-lived mothers (assume dod larger than today for those who are not already dead): pid I sex=:male AND mid IN pid I [dod-dob 90] This query is valid because IDs of mothers (mid) are of the same domain as IDs of individuals (pid), i.e. their comparison makes sense. However, in a normal data schema, it makes most sense to let all attributes for individuals be in relations that include the pid dimension. Thus, although pid and mid are both equivalently valid -77dimensions of the same domain, it can be said that the dimension pid is the primary dimension of the domain it belongs to, e.g. ID in 305. In general, one can therefore add to the metadata for domains a name of a primary dimension for any given domain.

One can further argue that it is more mathematically correct to specify output-domain as compared to output-dimension in a SDL definition. The example above would therefore be: SID I sex=:male AND mid IN ID I [dod-dob 90] There are other things that are elegant in this model, such as the definition of the "universe" that is needed for compiling expressions with negations. With outputdomains, the "universe" is simply a set (table, view or materialized view) with all values for the domain. For each domain, this definition set could be the union of all elements belonging to dimensions of its domain. This union can be maintained dynamically using materialized views. With regard to set-definition, this invention makes no distinction between a domain's primary dimension and the domain itself.

Thus the present invention defines a mechanism to associate a primary dimension to a domain and a relation that defines its "universe", e.g. two additional columns in table 305. By the same token, the default dimension prefix, mentioned in previous section, can be thought of as default domain prefix.

The main reason why the original language specification talks about an output dimension instead of output-domain is that it gives more flexibility (at the sacrifice of mathematical elegance). Consider for instance the problem of defining a set of all fathers that have more than two children based on the relations in Fig. 3: fidI count(pid) 2 This set is impossible to define based on the dimension pid, given only the relations in Fig. 3. In order for that to be possible, one would need to have a relation that defines a reciprocal relationship to that of 300, e.g. relation with children (pid, cid). Then the above query would become: pid I sex=:male AND count(cid) 2 In other words, all individuals that are male and have more than 2 children are selected. With a proper setup of SDL metadata it should be possible to define this reciprocal relation without any actual materialization of data using the fact that more -78- Sthan two dimensions can refer to the same SQL columns. Clearly, one can argue that Swith the use and definition of a primary dimension, a primary output-dimension is equivalent to an output-domain.

The initial reason for bringing up the discussion about primary dimension was the notion of a default output-dimension for nested sets. Given a dimension of a particular domain that is constrained on a nested set, the output of that nested set must N, be of the same domain. Now further assume that unless specified otherwise, the default output-dimension of a nested set is the primary dimension of the domain that ,i corresponds to the dimension that is constrained by the nested set. The example above could therefore be written as: pid I sex=:male AND mid IN [dod-dob 90] Furthermore, a default dimension prefix can be assigned to any primary dimension previously. Then the following two definitions are equivalent: objl::id size x OR hasobj2 IN {obj2::id I size z} and objl::id size x OR hasobj2 IN size z Thus, by default the nested set defines a set of objects of type obj2 whose size is larger than z. Note that the domain type of dimension obj I .hasobj2 is the same as that of dimension obj2. In Figure 15, it is shown how preprocessing steps can be used to implement default output-dimension, without modifying any of the earlier logic in the SDL compilation. Step 1501 identifies all nested SDL definitions. In 1502, for each nested set that does not explicitly define an output-dimension, a proper outputdimension is found. Finally, in 1503 the definition of the nested set is replaced with a new nested set that explicitly defines an output-dimension.

VIRTUAL DIMENSIONS Earlier, virtual relations (VR) were introduced as the basic data abstraction mechanism in the invention SDL system. The SDL system also supports virtual dimensions i.e. dimensions that are not mapped directly to RDBMS columns, but generated dynamically based on certain rules and the associated metadata.

-79- Therefore, together VD and VR provide data storage abstraction in the invention SDL language.

There are several categories of virtual dimensions in SDL. Here applicants only focus on virtual dimensions that are related to nested dimension drilling, briefly mentioned in the Nested Sets section above. Virtual dimensions and nested dimension drilling provides object-oriented "feeling" for the underlying data and uses cascadeddot-notation for dimensions like in path-expressions. Today, path-expressions are quite common in 00 programming languages, in OQL (Kim, W. "A model of queries for object-oriented databases", In Proceedings of the Fifteenth International Conference on Very Large Data Bases, August 22-25, 1989, Amsterdam, The Netherlands, P. M. G. Apers and G. Wiederhold, Eds. Morgan Kaufmann, 423-432; Cattell, R.G.G. Object data management: object-oriented and extended relational database systems, Addison-Weseley Publishing Company, Inc., 1994; and Stonebraker 1996), and more recently in XML languages (Chamberlin, "XQuery: An xml query language," IBM Systems Journal 41(4):597-615, 2002.). Pathexpressions have had many incarnations since they originated in (Mylopoulos, J. et al, "A language facility for designing database-intensive applications" ACM Trans.

Database Syst. 5(2):185-207, 1980) and have been extended to provide more sophisticated navigation capabilities (Kifer, "Querying object-oriented databases," In Proceedings of the 1992 ACM SIGMOD International Conference on Management ofData, ACM Press, 393-402 and den Bussche, J. V. and Vossen, "An extension of path expressions to simplify navigation in object-oriented queries," In Deductive and Object-Oriented Databases pp. 267-282, 1993). For the time being, the aim is simply to use VD as an alternative syntax for nested queries and therefore only their simplest form is considered here. Start with an SDL query similar to the one shown in Ex. pid sample IN sid I type "DNA" Ex. a7 With the VD notation, this same query can be written as: pid sample.type "DNA" Ex. a8 Notice the cascaded-dot-notation used in the representation of the sample. type VD. This dimension name is indeed generated by cascading the dimensions sample and type without their default prefixes. The full VD name with a prefix is equal to pid.sample.type. The translation of this query to SSDL is: SELECT DISTINCT a.<pid> FROM pid, sample.type AS a WHERE a.<sample.type> "DNA"; Ex. a9 Notice that the VR above is defined with a fingerprint that contains a VD. Before this SSDL statement is translated into SQL code it is expanded into the following code: SELECT DISTINCT a.<pid> FROM SELECT b.<pid>, c.<type> AS <sample.type> FROM [pid, sample] AS b, [sid, type] AS c WHERE b.<sample> c.<sid> AS a WHERE a.<sample.type> "DNA"; Ex. alO Thus, a VR that contains VD can be expanded into an SSDL statement that only contains concrete dimensions. Consider another expression that is very easy to generate with the support of nested dimension drilling and relates to the earlier genealogy query in Ex. a3: pid diag "stro.isch" AND father.diag "stro.isch" AND NOT mother.diag "stro" AND child.sex "male" AND child.diag "stro.isch" Ex. all This query is equivalent to Ex. a3 previously discussed.

An important feature of predicates on virtual dimensions in SDL is that they also apply for collections. An example of this is the child. diag VD which denotes any diagnosis of any child of a given individual. As with the standard treatment of predicates on concrete dimensions in SDL there is an implicit exist -81 quantifier. In some object-query language implementations predicates on collections are not permitted, however, other require the quantifiers to be specified specifically (den Bussche and Vossen 1993 and Bertino, et al., "Object-oriented query languages: The notion and the issues," IEEE Trans. Knowl. Data Eng. 4(3):223-237, 1992) or require predicates with membership functions (Stonebraker 1996).

The VD notation provides an alternative syntactic way to express constraints that otherwise would require nested set notation, thereby increasing the conceptual conciseness and intuitiveness of the language. Furthermore, this syntax provides more power when used in conjunction with binding variables than nested sets, because scoping rules do not allow binding variables to be visible outside of a set definition.

However, VD notation does not eliminate the need for nested sets, since expressions with predicates that involve aggregates and collection RD, may give unexpected results.

It is trivial to extend the cascaded-dot-notation for further levels of nesting, e.g. pid. father. father .yob would denote the year of birth attribute of the paternally related grandfather. Here yob is the last concrete dimension referred to in the VD name. To generate the virtual relation [pid, father. father .yob], the join [pid, father], [pid, father], and [pid, yob] are needed, where the joins take place between the RD and the corresponding PD. These relations specify a join-path that is used to generate the VR. In general, a join-path for a dimension can be generated by the following recursive algorithm, written as a function in a pseudo-language: VRlist JoinPath(Dim pd, Tokenlist Id, rd) if size(rd) 0 throw Exception; if size(ld) 0 and Dim(rd) in RelConcrDims(pd) return VR(pd, Dim(rd)); if Dim(ld) in RelConcrDims(pd) return VR(pd, Dim(ld)) JoinPath(Pdim(Dim(ld)), Null, rd); return JoinPath(pd, Id Head(rd), Tail(rd)); -82- The type VR is simply a virtual relation, and likewise VRlist represents an ordered list of such relations and the plus operator is used for list concatenation. The dot-separated dimension names are represented as token lists, and the function Dim (Tokenlist x) turns a token list into the corresponding dimension of type Dim. Similarly, the function RelConcrDims (Dim x) returns a set of all concrete dimensions related to the dimension x. The function Pdim (Dim x) returns the corresponding primary dimension based on the domain ofx. Finally, the function Head (Tokenlist x) returns the first token in the list whereas Tail (Tokenlist x) returns the list without the head. The reader can verify that JoinPath (pid, Null, father. father. yob) yields the correct result as shown above.

Similarly, if Pdim (Dim throws an exception if x is not a valid RD, then JoinPath an also be used to validate dimension names.

Figure 16 defines the process for determining if a dimension is a virtual dimension and finding (generating on the fly) the corresponding virtual relation for it.

The input 1601 is a record with multiple dimensions, including the output-dimension.

Note that it is necessary to treat many dimensions simultaneously for the purpose of determining if the "virtual record" is legal.

According to step 1602, names of regular dimensions override virtual dimensions (this allows virtual dimensions and the corresponding virtual relations to be materialized). Ifa dimension name is not recognized as a regular dimension it may be a virtual dimension. To determine this, a recursive algorithm that matches regular dimension names is used. The algorithm tries to match the left part of the given dimension name with existing dimension names that belong to relations with the corresponding output-dimension (taking into account dimension prefixes). Based on the match, the algorithm continues to try to match the rest of the dimension name by using dimension names that belong to relations with the primary dimension of the domain of the dimension found in a previous match, again taking into account dimension prefixes. This process continues until the whole input dimension name has -83been matched fully. If the input dimension cannot be fully matched, it is not a legal virtual dimension. This type of match is done for each input dimension in the record and information about all the dimensions that matched are collected, i.e. the path.

Step 1603 checks to see if the virtual dimension can be enclosed in the same record. For this to be true, the dimension path, up to the last dimension part, has to be the same as the dimension matching the so-called trailing parts in the virtual dimension names that belong to a record (relation).

If the virtual dimensions are all valid together, step 1604 generates the corresponding virtual relation by forming a join according to the path found earlier (see also examples).

Finally, in 1605 the dimensions are accepted and a structure for generating a common table expression for their relation is returned.

To clarify this complex definition, it is best to elaborate more on the examples above. Based on Fig. 3, the relation mid.dob (or mid->dob) is not recognized as a regular dimension. Also, dob is a name of another dimension (there is no prefix in this example) that can be joined with the primary dimension of its domain, e.g. pid, according to the definition specified in Fig. 16. This can be written in SSDL as: [pid, mid.dob] select A.<pid>, B.<dob> from [pid, mid] as A, [pid, dob] as B where A.<mid> B.<pid> Similarly, for an example where there is a default dimension prefix: [objl.id, objl.hasobj2.size] select A.<objl.id>, B.<obj2.size> from [objl.id, objl.hasobj2] as A, [obj2.id, obj2.size] as B where A.<objl.hasobj2> B.<obj2.id> Here, obj2.id is the primary dimension of the domain of the dimension objl.hasobj2.

Notice that obj2 does not appear between hasobj2 and size because of the prefix elimination.

A virtual relation based on multiple nesting, e.g. mid.fid.dob can also be defined as: -84- [pid, mid.fid.dob] select A.<pid>, B.<dob> from [pid, mid.fid] as A, [pid, dob] as B where A.<mid.fid> B.<pid> Here a virtual relation containing a virtual dimension is used to define the join, i.e. a recursive definition. This can be further expanded as: [pid, mid.fid.dob] select A.<pid>, C.<dob> from [pid, mid] as A, [pid, fid] as B, [pid, dob] as C where A.<mid> B.<pid> and B.<fid> C.<pid> As a remark, it is the part of the name in the virtual dimension prior to the last separator that defines the join needed for the corresponding relation. Each virtual relation that is defined in terms of virtual dimensions has to have a unique and unambiguous join-path.

Lets take another look at the earlier example: pid sex=:male AND [mid.dod mid.dob 90] Here the record-operator emphasized the fact that both mid.dod and mid.dob have to come from the same virtual relation. Thus, according to the present invention convention for noting the virtual relations in SQL: Vpid,mid.dob,mid.dod [pid, mid.dob, mid.dod] select A.<pid>, B.<dob>, B.<dod> from [pid, mid] as A, [pid, dob, dod] as B where A.<mid> B.<pid> Virtual dimensions can be implemented without changing anything in the parsing architecture of the SDL compiler if the metadata manager 17 has built in logic to handle them. This is somewhat a similar approach as current implementation uses to handle dimension aliases as discussed in the next section. Materialization of virtual relations can also be implemented into the metadata manager in connection with virtual dimensions, i.e. the view definitions above that are needed to declare the relations that contain the virtual dimensions can be materialized.

Support for virtual dimensions and default output-dimension of nested sets can be built into the client tools that are used for query composition. This support Applicants refer to as nested-dimension browsing and it will be discussed in a later section on the client components.

The Minimum Cursor Principle Table I. The join-paths for the individual dimensions in Ex. (a 12) with the VRs for each join-level shown in the upper part of the table. The lower part shows the extended virtual relations (CDC) and their cursor names for each join-level based on the minimum cursor principle.

-86- Dimension VR1 VR2 VR3 child.surname [pid, child] [pid, surname] child.mother.surname [pid, child] [pid, mother] [pid, child.father.surname surname] [pid, child] [pid, father] [pid, surname] Dimension CDCI CDC2 CDC3 child.surname a-[pid,child] b=[pid,surname] child.mother.surname a c=[pid,father,mother] child.father.surname d=[pid,surname] a c e= [pid, surname] Now consider an involved example in which is illustrated the use of virtual dimensions and how record-locking is extended into what applicants refer to as the minimum cursor or the minimum degree of freedom principle. Minimum cursor enforcement is intended to provide the semantics that are independent of the actual data layout, thus a part of the mechanisms in the SDL system that abstract data storage. There is a subtle difference between its implementation in automatic locking and explicit record-locking because of the possibility of ambiguity, i.e. multiple reference to the same dimension.

In automatic locking, the minimum cursor principle is implemented by grouping all VDs in a conjunct together that share parts of their join-paths. Within each VD group, all relations at each level are extended, for which the join-paths at lower levels are the same, into a CDC. This in only done if the merging is unambiguous, i.e. the same dimension is not used more than once in the VD group. If the same VD is repeatedly used in a VD group, the minimum cursor principle can nevertheless be applied without changing the semantics, for the left parts of the joinpath that are only based on non-collection relationships, i.e. relationships where the RD joins into a PD that has a PK constraint. In such instances, application of the minimum cursor enforcement is simply a performance issue. For any parts of the join- -87paths that are to right or at a level where there are collection relationships, independent cursors are used.

In explicit locking, with the record-operator notation, multiple use of the same dimension within a record-expression is always considered the same dimension, thus those join-paths all share the same cursors. Also, like in automatic locking, the relations at each level, which dimensions have join-paths that are equal at lower levels, are extended into a CDC.

First look at automatic locking through a fabricated example that involves multiple types of join-paths: pid I child.surname "Smith" AND child.mother.surname "Smith" AND child.father.surname "Smith" Ex. a12 The minimum cursor principle is based on the assumption that it is most natural and intuitive that a non-ambiguous conjunctive expression, like the one above, be interpreted in such a manner, that dimensions containing child in its name always refers to the same child. This would be the behavior expected from automatic recordlocking if the schema was denormalized and all these dimensions where from the same extended pid relation. Table I lists all the join-paths according to the JoinPath algorithm, for the dimension in Ex. (a 12), and shows the name of the cursors used at each level.

In this example, father and mother are both part of the singleton relation [pid, father, mother] whereas child is a collection that cannot be stored in the same concrete relation. Hence, child is not in the same CDC as father and mother. Notice that according to Table I, three cursors are used for the VR [pid, surname] since it occurs within three different join paths. The SSDL translation of Ex. (a12) is now easily written based on Table I: -88- SELECT DISTINCT a.<pid> FROM [pid,child] AS a, [pid,surname] AS b, [pid, father, mother] AS c, [pid, surname] AS d, [pid, surname] AS e WHERE a.<child> b.<pid> AND b.<surname> "Smith" AND a.<child> c.<pid> AND c.<mother> d.<pid> AND c.<father> e.<pid> AND d.<surname> "Smith" AND e.<surname> "Smith"; Ex. a13 Because both father and mother are singletons, the merging of [pid, father] and [pid, mother] into a single cursor, c, is simply a performance issue rather than a semantic issue. Clearly, this is not a trivial SSDL statement, however, its structure is relatively straightforward given Table I.

In order to hammer at the semantic meaning of Ex. (a 12), the reader might recognize that because of the minimum cursor principle, this query should only return the parents who have some "Smith" child whose father is a "non Smith" and whose mother is a "Smith". Assuming that there are no children born out of wedlock, this should be an empty set. In the discussion of Fig. 12, one saw how the record-operator can be used to minimize the scope of automatic record-locking. Likewise, one can use it to minimize the scope of the minimum-cursor enforcement. Thus, in the case of multiple marriages, the following query might give a non-empty set since independent cursors will be introduced for the children: pid child.surname "Smith" AND [child.mother.surname "Smith" AND [child.father.surname "Smith" Ex. a14 In Table II there is a similar analysis as in Table I for the query in Ex. (a14).

The SDL translation follows from the table.

Table II. The join-paths for the individual dimensions in Ex. (al4) and results for CDC and shared cursors. Because of the record-operators, the expression in Ex. (al4) is treated as three independent conjuncts.

L

-89- Dimension VRI VR2 VR3 child.surname [pid, child] [pid, surname] child.mother.surname [pid, child] [pid, mother] [pid, surname] child.father.surname [pid, child] [pid, father] [pid, surname] Dimension CDCI CDC2 CDC3 child.surname a=[pid,child] b=[pid,surname] child.mother.surname b=[pid,child] e=[pid,mother] g=[pid,surname] child.father.surname c=[pid,child] f=[pid,father] h=[pid,surname] The next example illustrates the treatment of ambiguous expression: pid child.diag "stro" AND child.diag "cancer" Ex. Here the same dimension is used multiple times in the same conjunct. The interpretation of this query, that is consistent with both normalized and denormalized data layout, is to find individuals that have some child that has been diagnosed with stroke and some child (same or different) that has been diagnosed with cancer. This is an example of where the exception in the minimum cursor implementation kicks in. If this must be one and the same child, then the nested notation must be used because the same dimension is being used multiple times in the same conjunct: pid child IN diag "stro" AND diag "cancer" Ex. al6 Notice how the above nested SDL notation suggests an alternative approach for translating queries with virtual relations, such as in Ex. (al2), i.e. first transform into nested SDL and then use standard SSDL translation: pidI child IN surname "Smith" AND AND mother IN surname "Smith" AND father IN surname "Smith" Ex. al7 This SDL translation is equivalent to a so-called star-join transformation (Bizarro, P. and H. Madeira, "The dimension-join: A new index for data warehouses," in XVI Brazilian Symposium on Databases (2001) and Pirahesh, H. et al., "Extensible/ rule based query rewrite optimization in Starburst," in Proceedings of the 1992ACM SIGMOD International Conference on Management of Data, ACM Press, 39-48, 1992) that is used in some RDBMS optimizers to obtain better performance on starschemas. This is however not guaranteed in all RDBMS and there are examples where this type of rewrite will give worse performance. Ideally, the SQL optimizer should be able to recognize when and how it is optimal to rewrite Ex. (al3). In expressions with the ALL keyword or aggregate operators, the rewrite shown in Ex.

(a 17) may not be permissible because of semantic difference for collection relationships. Examples of such are considered later, but first an example where only part of a conjunct is ambiguous: pid father.diag "stro" AND father.diag "cancer"

AND

child.diag "stro" AND child.diag.date "2000-01-01" Ex. a18 In this case, automatic locking will treat this query as if it was equal to: pid father.diag "stro" AND father.diag "cancer"

AND

[child.diag "stro" AND child.diag.date "2000-01-01" Ex. a19 In other words, child always refers to the same child because the child dimensions are unambiguous whereas father. diag refers to independent diagnosis of the father.

a) C4 Cc, -91 Dimension VR1 VR2 child.diag [pid, child] [pid, diag] child.diag.date [pid, child] [pid, diag.date] child.sex [pid, child] [pid, sex] child [pid, childl] Dimension CDC1 CDC2 child.diag a=[pid,child] b=[pid,diag, diag.date] child.diag.date a b child.sex a c [pid, sex] child a Table III. The join-paths for the individual dimensions in Ex. (a20), corresponding CDCs and shared cursors.

Aggregates can be used in combination with virtual dimensions like concrete dimensions, however, they require a special attention as mentioned above. For instance: pid I COUNT(child.diag) 1 WHERE child.diag "cancer" AND child.sex "male" GRBY child Ex. This query finds individuals that have male children with more than two cancer diagnoses. Note the additional GRBY clause. If grouping had been omitted, this query would find individuals for which the total number of cancer diagnosis for all sons is more than one. Particularly noteworthy in this example is that within the recordoperator it is possible to use simultaneously virtual dimensions that have join-paths that end in different concrete relation. This is an important property in order to support abstraction of storage. As Table III shows, sex and diag belong to different CDCs. Therefore, one can say that for virtual dimensions of the type pid. child.*, applicants have "extended" the virtual relations as compared to VRs for pid.* which are the CDCs. These extended virtual relations (EVR) will be discussed further later.

With regard to the behavior of the aggregate, in Ex. (a20), this extension does not matter because sex is a singleton. However, if sex had been replaced with a collection such as child. drug, the COUNT aggregate operator would yield "unexpected"results due to a multiplication effect from the join. A possible remedy to -92this problem is to transform p(child. drug) into child IN p(drug) This solution does however not work in general, e.g. cases where predicates depend on multiple dimensions from multiple join-paths. Rather, applicants suggest to raise a warning (through syntax and semantic aware GUI support) or simply to reject virtual relations where the join-path goes through a collection relationship, unless this joinpath is part of the join-path of the aggregated dimension itself. Likewise, recordexpressions with aggregate operators on multiple join-paths are not accepted if the join-path goes through collection relationship. This is done to avoid "unexpected" behavior from predicates with aggregate operators in EVRs. As an example, the following would be flagged: Spid COUNT(child.diag) 1 WHERE child.drug "t-PA" Ex. a21 whereas Ex. (a20) is valid and similarly the following: pid COUNT(child.diag) 1 WHERE child IN drug "t-PA" Ex. a22 Before ending this section and the discussion on the minimum cursor principle, it is recognized that so far only expressions that have a single conjunctive term have been presented. More complex expressions can however always be rewritten, by applying the distributive law or an OR-distribution, into CNF (Ramakrishnan, R. and J. Gehrke, Database Management Systems, 2nd Ed., McGraw Hill, 2000). For automatic locking, each conjunct can be considered separately as described above, hence ambiguity would be determined for each conjunct. Within record-expressions, however, OR-distribution is not applied since it can change the result of aggregate predicates.

Dimension aliases Dimension aliases can be defined in the SDL metadata, i.e. a dimension can be recognized under different names and thereby different dimension names can produce the same dimension fingerprint. Note that the same table column can also be associated with more than one dimension, however, that is slightly different than -93dimension aliases since they apply also to virtual relation creation on SSDL level. The metadata and the metadata manager must be modified such that dimension names and its aliases produce the same fingerprint. There are many ways to do this, however, to avoid combinatorial complications from multiple dimensions and multiple aliases when searching for corresponding relations, it is best to use one main dimension name for lookup of the relations. Known techniques for such handling of aliases are suitable. Later discussed however is how aliases can be useful in mapping objects to relations (not to be confused with object-relational extension in the next section) as well as for federation.

Object-relational extensions Virtual dimension are closely related to objects and data encapsulation. SDL is easily extended to use object-relational features such as those provided in Informix UDB (Illustra), DB2, and Oracle9i An SDL domain can be defined in terms of extended data types (objects). Similarly, methods that belong to the extended data type can be made accessible visually (in a tree browser) as well as syntactically in the same manner as attributes and virtual dimensions. Furthermore, operators can be overloaded and assigned to domains in the same manner as in languages such as C++ (Stroustrup, The Programming language, 2nd ed. Addison-Wesley, 1991).

As an example, the operator in SDL can be overloaded for non-hierarchical data types.

Next are examples with a data type called "segment" that has the methods start(, stop(, and chrom( for chromosomes in a biotech example. Visually, these methods can be shown as leaves under the dimension-node seg, with parenthesis appended, and possibly with information on input parameters. They can be used in an SDL statement as: gene| func=: "GPCR" AND seg.chrom()=1 AND seg.start()>10000 -94- Similarly, if an object-relational method for evaluating overlap between two segment objects exits, the following query could be used to find if any "non-silent" SNPs a single nucleotide polymorphism in DNA that impacts RNA transcription of a gene) overlap GPCR genes: genel func "GPCR" AND $x seg AND SIZE({snp type "nonsilent" AND seg.overlaps($x) Notice that the binding variable, is of object-relational type. Similarly, it is possible to write method i.e. to refer to a method through the binding variable. The SIZE keyword in SDL is simply used to evaluate the size of the nested set as mentioned previously.

An important issue with regard to object-relational methods is that they do not require any meta-data management in the SDL system apart from the domain definition. Registration and definition of methods can simply take place at the RDBMS level since the SSDL translator can simply append the methods to the corresponding column names.

XML data and SDL queries Next shown is how SDL can be used to in connection with XML data. By proper shredding of the XML data tree into relations, SDL can be used to search for objects such as books, based on the attributes that have been defined in terms of XML.

Below is given an example based on a document type definition (DTD) found at the XML consortium, www.w3c.org. Consider the following definition of a book: <!ELEMENT book (title, (author+ I editor+ price <!ATTLIST book year CDATA #REQUIRED <!ELEMENT author (last, first <!ELEMENT editor (last, first, affiliation <!ELEMENT title (#PCDATA)> <!ELEMENT last (#PCDATA)> <!ELEMENT first (#PCDATA)> <!ELEMENT affiliation (#PCDATA)> <!ELEMENT price (#PCDATA)> There are many ways to map XML to relational data structures see reference F. Tian et.al. "The Design and Performance Evaluation of Alternative XML Storage Strategies", ACM Sigmod Record, vol. 31(1), 2002). This discussion is not a formal discussion on how XML can be mapped to relations, but more to serve as examples of how a DTD can be mapped to a relational structure and corresponding dimensions in order to demonstrate how SDL can be used with XML. For instance one could use the following relations in one embodiment: (bookid, title, year, price), (bookid, author.last, author.first), (bookid, editor.last, editor.first, editor.affilation) With these virtual relations a user could for instance define the following SDL queries in order to search for books in libraries: bookid [title="*SQL*" AND year>1900] AND (author.first="John" OR editor.first"John") As mentioned earlier, the record operator could be omitted if it is assumed that each book only has one title. The SDL system could automatically enforce the record locking (for increased efficiency) if that relation has a primary key or if title and publisher are declared to belong in a dimension structure. Another example from this is worth considering, where the fact that a book can have many authors is used in the query: bookid [author.first="John" AND $x:=author.last] AND [author.last=$x AND NOT author.first="John"] This query would find all books that have two or more authors with the same last name and one of these authors with the first name of John. In a typical setup where the system is aimed at providing the user with a mechanism to find books, the user would not have to specify the output-dimension, book.id, or the set curlybrackets.

-96- The schema above is constructed based on the fact that queries will be made to select books. Thus, it has been denormalized in some sense to reflect that. This is the inherent nature of XML, i.e. the hierarchical structure of XML means that one has to make a choice. It is, however, well known that proper relational database design gives a lot of flexibility with regard to update of data, speed and ability to pose arbitrary queries. Here Applicants suggest different relations for the above DTD, e.g.: rl (book.id, book.title, book.year, book.price), r2 (book.id, book.author) (author.book, author.id), r3 (book.id, book.editor) (editor.book, editor.id), r4 (author.id, author.first, author.last), (editor.id, editor.first, editor.last, editor.affilitation) Applicants have labeled the relations to indicate names of possible SQL relations/tables that store the data. Notice also that Applicants map two distinct virtual relations to the tables r2 and r3, i.e. different dimension combinations. The reason is that these are relations that store relationships between objects and the purpose of mapping them two times will become clear as this discussion continues. In this schema, assume that book.author and author.id are of the same domain and that author.id is the primary dimension. Likewise, book.editor and editor.id are of the same domain. The following queries are now possible: book::id [title="*SQL*" AND year>1900] AND author IN first="John" book::id I [title="*SQL*" AND year>1900] AND author. first="John" With the alternative mapping of r2 a set of authors can now be easily defined, e.g.: author::id first="John" AND book IN title="*SQL*" One can argue that to make these queries even more readable, the dimension/attribute author.book should be called author.haspublishedbook etc. Proper naming of dimensions is an important subject but is outside the scope of this discussion.

-97- It is now worthwhile to show an example that uses virtual dimension notation, as compared to nested sets.

author::id I first="John" AND book.title="*SQL*" All of these variants are automatically handled by the SDL systems from the same relations, given that it supports default dimension prefixing and output dimension as well as virtual dimensions.

As a final example of the flexibility of SDL in terms of short and yet powerful queries on XML data that has been mapped to relational structure consider the problem of finding all authors that have published more than ten books or have been the single author of a book: author::id I count(book)>10 OR book IN {count(author)=1} The aim of this section was to show that XML data can be mapped to relational data structures, which are then easily used through SDL to search for various types of objects. In the next section, the related issue of the use of searching for objects that have been mapped through Object-Relational mapping to a relational schema is considered.

Objects, relational mapping and SDL queries Next, consider how SDL can be used as a query language for objects. The subject is very similar to the mapping of XML, however, this discussion will also consider inheritance of objects. The aim with this discussion is not to cover this subject in general, but to give a simple example in order to demonstrate the flexibility that the SDL language and its metadata provide to implement a powerful object-query language. The following example will be very similar to the XML examples, however, the below will use object-inheritance to implement authors and editors, i.e., both authors and editors are object classes inherited from the person class. Some basic class definitions could be the following (in a Java/C++ like syntax): Class Person String last, first; Class Author Person Book books[]; -98- Class Editor Person Book books String affiliation; Class Book String title; Integer price, year; Author authors Editor editors[] Object oriented languages provide a mechanism to cast objects, either using object-widening or object-narrowing. By using multiple dimensions for the same table columns (dimension aliases) and multiple domains, it is possible to provide this support in an elegant manner in SDL. Consider the following relations that Applicants use to map the above classes: rl :(book.id, book.title, book.price), r2 (person.id, person.first, person.last, person.objtype), r3 (book.id, book.authors) (book.id, book.persons) (author.books, author.id) r4 (book.id, book.editors) (book.id, book.persons) (editor.books, editor.id) (author.id, author.first, author.last, author.objtype="author") (person.id, person.first, person.last, person.objtype="author") r6 (editor.id, editor.first, editor.last, editor.affilitation, editor.objtype="editor") (person.id, person.first, person.last, person.objtype="editor") Worth noting here is that relations/tables r5, that store the attributes of the author objects, and r6 that stores the attributes of the editor objects, are also mapped to the dimensions that represent the attributes of the person objects. Furthermore, for r6, the second dimension mapping only uses three of the 4 columns, since the person object does not have any attribute for affiliation. The second dimension mapping of table r3 and r4 are to support narrowing. An important issue here is that the domains of person.id, author.id, and editor.id are different, although they will have to use the same underlying data type. These dimensions are also defined as the primary dimensions for their corresponding domains.

Like usually this is best explained through examples: person::id first="John" This query finds all person objects with the first name of John. In order to evaluate this query the virtual relation of [person.id, person.first] is used. Given the metadata definition defined above, it will find all objects that have been put into relation r2, -99and r6. To find all the objects for authors with the first name of John the two following definitions would be possible: person::id first="John" AND objtype="author"} author::id fisrt="John" An example of a query that uses the equivalence of widening would be the following: book::id title="*SQL*" AND person IN person::id first="John" AND objtype="author" Note that person refers here to either an author or an editor. To define object casting that is equivalent to object narrowing, one embodiment maps two dimensions to the same column in relations r5 and r6, (author.id, auth2pers) and (editor.id, ed2pers) respectively. Note that even though the two dimensions in each relation map to the same column, they must be of different domains. This would for instance enable the following query: book::id title="*SQL*" and author IN author::id auth2pers IN person::id first="John" It is not necessary to specify the output-dimensions of the nested sets in the above examples, however, it is done here to emphasize the different types of object from each set. Alternatively, this could have been written using virtual dimension notation: book::id title="*SQL*" AND author.auth2pers IN person::id first="John" This concludes the discussion of how SDL can be used to define sets of objects and how multiple mapping of dimensions to the same tables can be used as a mechanism to implement transparent yet efficient object-inheritance and objectcasting in conventional RDBMS.

SDL federation The SDL system 100 provides several mechanisms to deal with federation of its metadata 17. As such, the SDL system does not implement any specific techniques to federate queries but relies on federated RDBMS, such as DB2 DiscoveryLink.

-100- Thus, SDL delegates the optimization to the underlying database system. However, the SDL system provides tools to setup a SDL server that is built from other SDL servers, local or remote, by using metadata from the metadata managers of these servers to define the metadata in the new server. A unique feature of the SDL system is that it allows this to be done not only when the new SDL server is defined, but also session dependent for each user that accesses the SDL system. Furthermore, each session can provide its own storage manager, for storage of results and other user dependent data.

There are two main cases where federation ofmetadata is desirable, i.e. when multiple database instances store equivalent data or when data of different nature are stored in separate servers and relationships between the data enables queries that refer to multiple servers simultaneously. The SDL system 100 provides two different mechanisms to implement federation. The former is referred to as static federation and the latter as dynamic or session dependent federation.

Static federation If two or more SDL servers exist they can be combined using federation technology. The SDL data management tools provide options to define how such federation takes place, i.e. which server and how their metadata is combined. Static federation is achieved with merging and prefixing multiple metadata schemas into a single union schema as illustrated in Fig. 17.

In a typical scenario, a new SDL server is generated on a DB2-DiscoveryLink (DL) server. One or more external (can be internal as well) SDL servers 1701 are specified for federation. These external SDL metadata schemas will have a corresponding link schema 1702 on the DL server. Such link schemas 1702 are implemented by creating nicknames to the necessary metadata tables 17 in the external SDL servers 1701 (instead of DL nicknames, Oracle DB links can be used as well).

The other step in the federation is to combine the dimensions from all the SDL servers 1701, i.e. all the link schemas 1702, into a single SDL metadata schema 1703, the union schema. Often a straightforward copying is not sufficient because there can 101 be conflicts in dimension names between the servers or the dimension attributes may need to be organized in a new manner given new dimension attributes from other servers. To achieve this, the SDL system provides two main mechanisms, i.e.

dimension prefixing as well as dimension and domain aliases.

Consider the case where there are two servers, both set up for genes. One stores basic annotation data, and the other stores information on gene expression measurements. In server A, there are the following relations: (gid, start, stop, chrom, name) (gid, go_term) and in server B there is the following: (ugid, name) (ugid, level, expr) (expid, tissue, date, condition) One may now want to combine these dimensions into a new SDL server by specifying a prefix on all the dimensions, e.g. A and B. This would automatically generate the following dimensions: (A.gid, A.start, A.stop, A.chrom, A.name) (A.gid, A.gene_ontology) (B.ugid, B.name) (B.ugid, B.level, B.expr) (B.expid, B.tissue, B.date, B.condition) Furthermore, one can define dimension aliases (gid A.gid) and (gid B.ugid). Note that defining aliases is similar to mapping the same relation multiple times, as in some of the previous examples such as in the section on object-relational mapping, however, it is more convenient if the same dimension appears in multiple relations such as is often the case for primary dimensions. Now query both of these databases, as an example: gid A.name B.name This will generate all genes with the same name in both databases. Similarly, consider: 102gid A.chrom=:autosom AND A.go_term=:gpcr AND [B.level>0.9 AND B.expr IN {tissue="brain" AND date>2000}] Both of these queries above would actually result in a federated join. For instance the first query would generate: select distinct TA.gid_col from TA_gid_name as TA, TBugid_name as TB where TA.gid_col TB.ugidcol; In the above query, the table names (nicknames more appropriately) reflect the location of the underlying data tables. If in addition one generated two aliases, (name A.name) and (name B.name), then the following query gid name="APO*" would generate: select distinct TAB.gid from (select TA.gid_col, TA.name_col as name from TAgid_name as TA UNION ALL select TB.gid_col, TB.name_col as name from TBugidname as TB) as TAB where TAB.name like "APO%"; Clearly, prefixes and aliases can be used as a simple intuitive mechanism to organize federated SDL metadata. In addition, as brought up in previous sections on pivoting dimensions, relations (in particular those with aliases) could be included into the union schema by introducing an additional server or database dimension whose value reflects the origin of the data tuples. This is easily implemented through a view mechanism on the nicknames and would for instance allow queries such as: gid I name "APO*" AND server gid I A.name "APO*" A.gid I A.name "APO*" Dynamic federation A novel federation feature in the SDL system 100 is referred to herein as dynamic or session dependent federation. What is meant is that users can specify the federation set-up at runtime. Thus each session in a federated SDL server can have different federation set-up. This capability opens up many new possibilities. With regard to application development, dynamic or session dependent federation setup will 103 allow applications to determine at runtime which SDL compatible data sources it federates. Thus, for instance, session dependent application data can be used within the SDL query tool, as long as the application data has a relational API and the application provides SDL metadata for that data. A relational API to the application data is most easily achieved by storing the given application data in a relational database but relational wrappers can also be defined for data that does not reside within a relational database. Unlike in the static federation, an application that uses dynamic federation defines its metadata by providing the SDL server a handle to one or more SDL schema feature objects, prefixes and aliases (as in the static federation).

Figure 18 shows a preferred embodiment of an architecture to implement dynamic federation in SDL. An application 1821 with an embedded SDL client 1849 logs into an SDL server 1800 via 1801. Next it (1821) instantiates 1802 an SDL schema feature object 1823 in a local application server 1850 that maps the application data. This data can be persisted in a local SDL server, data in an arbitrary relational database, flat files or even in applications main memory, if appropriate wrappers are used. Once the SDL schema object 1823 has been created (or requested from a local SDL server), it is registered 1803 on the remote SDL server 1800. The metadata session object 1881 on the remote SDL server 1800, which now contains two SDL schema objects 1825 and 1826 (possibly more), lets the remote federated features register themselves 1804 such that the federation engine in the RDBMS 1851 can connect 1805 to the federated data 1899, e.g. if the federated data resides in a RDBMS. Once this has taken place, the SDL client 1849 can start issuing queries 1806 to the SDL system 1800. The SDL system 1800 compiles the SDL statement to an SQL statement 1807 based on the session's total metadata 1881. Nicknames or aliases that have not yet been defined, but are referred to in the SQL statement, are generated on the fly 1808 by the metadata-manager 17. Finally, the SQL statement is executed 1809 and federated appropriately 1810 by the federation mechanism in the RDBMS 1851.

SDL Client Components 104- This invention describes several client software components that can be used to facilitate the creation of SDL queries. Figures 19A and B show a composite query O tool comprised of several independent components. According to Fig. 1, the tool in Figs. 19A and B can be divided into a data explorer 1900 and 1901 and a querycomposer 1903 and 1904. In 1900 one sees a tree-browser that provides an overview Sof the meta data in the system. In this example, it shows the dimensions presented in 'I hierarchical manner. It could also present the dimensions beneath the corresponding Odomains or the SQL relations that they belong to or any other system that may be N useful for the user to group and classify the dimensions. The window in 1901 shows the dimensions that are related to the selected dimensions in 1902. This window can also be configured to show all the SQL relations that the selected dimension exists in.

An SDL syntax aware editor is shown on the right side of Figs. 19A and B. It is split up into the editing part 1903 that shows the dimensions and a description part 1904 that shows the descriptions of the dimensions that are stored in the metadata. Finally, shown are buttons 1905 for launching the SDL query and a list box 1906 showing information about the sets that have been generated with SDL queries.

For enumerable dimensions, the data explorer 1900 has the unique feature of allowing the user to drill into the domain values that are stored in a corresponding hierarchy, e.g. the gene ontology classification scheme 1907. This enables the user to drag and drop into the editor both the dimension and the corresponding condition on that dimension 1908. The description view 1909 shows the corresponding statement with the description taken from the hierarchy table, e.g. 307 in Fig. 3.

The SDL invention users are not required to think about the tables that store the data. However, they need to understand when they can apply the record operator on relations. Figures 20A and B show how the composite tool supports that. For the selected dimension, 2000, the view 2001 shows that the attribute "chromosome" exists in a relation with the attribute "gene-symbol". Hence, the user can apply the recordoperator as shown in 2002. The editor also warns the user if he does not enclose dimensions that exist in a relation into a record operator 2004, e.g. with underline marks. This does however, not necessarily have to be an error (see the discussion of automatic enforcement of record operator). As shown in Figs. 21A and B, the editor 105 also has syntax aware support for the insertion of SDL keywords 2100 and dimension and domain values insertions 2101.

Figures 22A and B show how the editor 1903, 1904 can be configured such that the user does not have to specify either the output-dimensions or the curlybrackets of the set. The expression in 2200 is an example of this and notice that the output-dimension in the nested set can also be omitted if it is the same as the default output-dimension. In this example, the default output-dimension is GID as shown in 2201. The composite query tool provides a Venn tool 2202 that allows sets of the same type to be analyzed. In one embodiment, the Venn tool is generally an automatic SDL query generator.

The composite query tool provides a very simple mechanism to specify a report for the elements in the query sets. Figures 23A and B show the report outputspecification in the preferred embodiment. Actually, the output specification tool is a relation definition tool and the relations generated with it could be registered in the SDL system metadata. However, in the present invention, applicants only consider it for the purpose of generating relations that can be joined with the SDL query sets, in order to provide more information on the elements in the sets. The output specification tool provides drag and drop support and the user selects the dimensions he wants to see together, e.g. GID and TEXT.DESC 2300, and places them in the report, 2301. By a mouse double click on any set in the list box, 1906, a report corresponding to the set and the output spec opens up, 2302.

As shown in Figs. 24A and B, the output specification (report builder) also supports the creation of columns with aggregate operators 2400 (the standard SQL aggregate operators) as shown in 2401. Furthermore, the output specification tool allows the user to specify whether the relations that contain the dimensions should be joined with the set using outerjoin or regularjoin 2402. Also, it allows the user to specify if all the dimensions in the report have to come from a single relation or not.

If strict is selected 2403 the system will complain if more than one relation is needed to cover all the dimensions specified in the output specification. Notice that the output specification is to specify an operation, i.e. the report generation, which is independent of the SDL set definition, although both depend on the definition of an output- 106dimension. As mentioned before, SSDL is ideal for the purpose of generating instructions to create a report.

Even though the syntax-aware editor in conjunction with the drag and drop behavior of the metadata tree provides a user-friendly interface, it is even easier to issue queries through specially designed dialogs. Because of the simple and sparse nature of the SDL language, it is relatively easy to create a drag-and drop driven dialog builder in which users can create their own dialogs without any programming effort. In the preferred embodiment of the composite query tool, applicants describe a dialog builder as shown in Figs. 25A and B. The user can simply drag the dimensions from tree, 2500, and drop them onto a canvas. This will create an input field for the corresponding dimensions in the dialog. The user can then build expressions with Boolean logic by connecting the input fields with either AND or OR as in 2501.

Furthermore, the user can enclose input fields from the same relation into a record operator by labeling the appropriate dimensions with a tick mark and then locking them together 2502.

The dialogs can be locked and launched such as from the resource tab. The dialog then prompts the user for input 2503 for all the fields in the dialogs. The inputs will then become constraints on the corresponding dimensions, e.g. disease type of a lod score in 2503. For enumerable dimensions the dialog even supports browsing of the corresponding domain hierarchy 2504. Once the user presses the query button in the dialog, an SDL query is launched and the result set appears in the list-box 2201, like the sets defined from the editor. A particularly nice feature of the dialogs in the preferred embodiment is that if the user does not complete all fields, the dialog will generate a reduced expression. The rules resemble the way NULL is treated in Boolean logic in SQL, e.g. the SDL expression {el AND unknown AND e3} will become {el AND e3} and {el AND unknown OR e3} will become (el OR e3}.

Since the expressions that the dialogs support are usually relatively simple SDL expressions, the reduction of the expression based on undefined fields does not cause confusion. This type of reduction mechanism can also easily be extended to more complex expressions that involve brackets. The rules above can be applied within each bracket, and only if the entire expression within the bracket results in unknown is 107the bracket replaced with unknown. In many ways, the dialogs can be considered as SDL setnames (formulas) with parameters (see the SDL syntax specification). Indeed, the present embodiment stores setname formulas and dialogs in a similar manner.

Figures 26A and B show how the present invention provides a template builder for a universal resource locator (URL), i.e. a tool for defining web-based actions.

Other actions, such as Windows based OLE actions could also be defined in a similar manner. The URL templates can be defined in terms of fingerprints of one or more dimensions 2601 The URL can then be defined in terms of symbols indicating the desired dimensions 2602. Furthermore, a description 2603 can be assigned to the given URL template. In SDL reports, the user can then activate URL templates that are compatible with the dimension fingerprint of the given report. By selecting a specific line in the report the user can then choose a template 2604 to launch, which is then populated by the corresponding dimension values in the report 2605. As a remark, it is noted that URL templates can also be based on domain fingerprints as compared to dimension fingerprint.

Nested dimension drilling is a GUI support feature for creating nested SDL queries or SDL queries with virtual dimensions. Figure 27 shows an exemplary data tree-browser in an SDL query composer 1903, 1904. Here it is assumed that the metadata has been defined in such a way that pheno.id 2701 is the primary dimension of the domain shared by pheno.id 2701, pheno.gen.mid 2702 and pheno.gen.fid 2703.

Other dimensions, such as pheno.dem.dob 2704, are of different domain, a domain for which no primary dimension has been defined. Therefore, one sees no symbol or indication for drill-in (plus or minus) in 2704 as is seen in both 2702 and 2703. Once the user chooses to drill-in on pheno.gen.fid 2703, the present invention provides two options: to drill-in for generating nested sets 2705 or to drill-in to generate virtual relations 2706. In Fig. 27 it is assumed that the user chose 2706 to select a virtual dimension. Regardless of which of the two selections the user makes, he will see a new dimension tree fold out with all the dimensions that belong to relations with the primary dimension of the domain that 2703 belongs to. Thus, dob 2707 denotes the virtual dimension pheno.gen.fid.dem.dob, which would be the dimension that is inserted into the SDL 2708 if it is dragged-and-dropped into the editor. In contrast, 108had the user chosen to drill-in through 2705, a drag-and-drop would have resulted in 2709 in the SDL editor.

Since the tree browser can become a little too crowded with these drill-in indications (2705 and 2706), one embodiment of the present invention makes them optional. The user can press a button to hide them from the tree. With Fig. 27 in mind, it should not be too hard to envision an extension of the tree browser where other features of dimensions are made available in a similar manner as the above drillin indications. Such features can be object-relational methods, as mentioned earlier, supported operators etc. Indeed, the domain value browsing 1907 is a special case of such support provided in an exemplary embodiment of the invention. Presentation of all of these features could be customizable. Similarly, the tree-view could be customized to show only the dimensions that are related to specific primary dimensions, e.g. with events from the context sensitive SDL editor. All such support can make the query building process easier for the user.

EXTENDED VIRTUAL RELATIONS The previous discussion on the record-operator in section "Minimum Cursor Principle" outlined how the virtual relation could be extended for virtual dimensions from multiple CDCs, e.g. [pid, father. diag, father. sex] where diag and sex are stored in different CDC's. Similarly, the above alluded to the fact that this same extension could be done for concrete dimensions as well, although it does not provide any additional simplifications to queries since the SDL language already has a built in implicit join based on the output primary dimension, as mentioned previously. Such extension is nevertheless valuable for virtual relations in SSDL and for object (PD) report generation. Here extended virtual relations (EVR) are explained through a few examples but an elaborate definition is not provided since EVR's mimic explicit record-locking and the minimum cursor principle.

Working with dimensions and relations from earlier examples, the first example is very simple and presents how an EVR is generated with SSDL from two CDCs: 109- [pid, yob, sex, diag, diag.date SELECT a.<pid>, <yob>, <sex>, <diag>, <diag.date> FROM [pid, yob, sex] AS a, [pid, diag, diag.date] AS b l WHERE a.<pid> b.<pid>; Ex. a23 0 In essence, an EVR abstracts how dimensions from several CDCs are combined together into a single virtual relation, e.g. singleton information such as sex is stored Sin a different concrete relation than the collection diag. Similarly, one can write [pid, sex, diag, drug], i.e. refer to dimensions from three CDCs. There are however two issues that require special attention when multiple CDCs are combined.

First, when an EVR refers to dimensions from more than one collection CDC, inherently there will be "multiplication effect" on parts of the tuples. This is a consequence of presenting data structures that are more naturally presented as some kind of a hierarchical structure XML or the CORBA IOP format) than as a relational structure. The relational form is however more easily viewed in text editors or spreadsheets and it is directly applicable to further processing using relational languages. As mentioned in the "Minimum Cursor Principle" section, this "multiplication effect" from multiple collections does however impact aggregate operators. Singleton clusters are however invariant to equi-join on a PD with another singleton cluster.

Second, equi-join of multiple CDCs could potentially lead to "loss of data" if tuples of a given PD were absent in any of the CDCs. The definition of virtual relations does however guarantee that all the values of a PD in the domain definition set has a representative tuple, even though it may only contain NULL values for dimensions other than the PD. Hence, the implementation of EVR in Ex. (a23) which uses equi-joins between the CDCs is "non-lossy" by definition. In practice, due to performance reasons, it may not be desirable to implement EVR based on that assumption, but rather build the EVR from concrete relations that only contain non NULL tuples. Applicants define a virtual relation based on a strict fingerprint as: -110- [dpi d, db [dpi d, db WHERE =NULL OR db! =NULLI Ex. a24 Here a WHERE clause has been introduced in the EVR that has equivalent behavior as in the record-operator in SDL definitions. Now the EVR in Ex. (a23) can be defined in terms of CDCs with strict fingerprint: [pid, yob, sex, diag, diag.date I SELECT a.<pid>, b.<yob>, b.<sex>, c.<~diag>, c.<diag.date> FROM (SELECT <pid> FROM [pid]) AS a LEFT JOIN (SELECT <pid>, <yob>, <sex> FROM [[pid, yob, sex)]) AS b ON a.<pid> b.<pid> LEFT JOIN (SELECT <pid>, <diag>, <diag.date> FROM [[pid, diag, diag.dateJ]) AS c ON a.<pid> Ex. This definition is "non-lossy" since it will minimally generate a single row for each pid in the domain definition set.

Now consider an example of an EVR with an aggregate operator and a constraint on the PD: E pid IN I diag "stroll yob, sex, COUNT(diag), cliag.clate SELECT a.<pid>, b.<yob>, b.<sex>, COUN T(c.<diag>), c. <diag. date> FROM (SELECT <pid> FROM (SELECT DISTINCT <pid> FROM [[pid, diag]] WHERE (<diag> LIKE llstro*1I) AS a LEFT

JOIN

(SELECT <pid>, <yob>, <sex> FROM [[pid, yob, sex]]) AS b ON a.<pid> b.<pid> LEFT JOIN (SELECT <pid>, <diag>, <diag.date> FROM [[pid, diag, diag.date]]) AS c ON a.<pid> c.<pid> GROUP BY a.<pid>, b.<yob>, b.<sex>, c.<diag.date>; -Il1- Ex. a26 Predicates can be applied directly where the output dimensions in the EVR are listed or in a WHERE clause as shown earlier. The domain definition relation [pid] in Ex. (a25) has been replaced with the SSDL translation of the constraining set for pid. Observe that the grouping is implicitly determined by the non-aggregated output-dimensions. Most importantly, the grouping is by pid and diag. date.

Because both yob and sex are singletons, they yield no further segregation than pid. If drug had been included as well, i.e. a dimension from another collection than diag, the aggregation would have been "incorrect". Proper aggregation can nevertheless be achieved with the aid of nested EVR notation: pid, yob, sex, COUNT(diag), diag.date COUNT(drug) I WHERE pid IN diag "stro" (SELECT a.<pid>, b.<yob>, b.<sex>, c.<COUNT(diag)>, c.<diag.date>, d.<COUNT(drug)> FROM (SELECT <pid> FROM (SELECT DISTINCT <pid> FROM [[pid, diag]] WHERE (<diag> LIKE "stro*") AS a LEFT JOIN SELECT <pid>, <yob>, <sex> FROM [[pid, yob, sex]]) AS b ON a.<pid> b.<pid> LEFT JOIN (SELECT <pid>, COUNT(<diag>) AS <COUNT(diag)>, <diag.date> FROM [[pid, diag, diag.date]] GROUP BY <pid>, <diag.date>) AS c ON a.<pid> c.<pid> LEFT JOIN (SELECT <pid>, COUNT(<drug>) AS <COUNT(drug)> FROM [[pid, drug]] GROUP BY <pid>) AS d ON a.<pid> d.<pid>; Ex. a27 Notice how the scope of the aggregation and the grouping is within the square brackets. Therefore, this "report" returns demographic information on individuals, their number of diagnosis, as a function of date, as well as their number of drug -112prescriptions. Similarly, it is possible to use RD and nested EVR notation to create nested reports: pid, sex, COUNT(drug) child sex, COUNT(drug) I SELECT a.<pid>, b.<sex>, c.<COUNT(drug)>, d.<child> c.<diag.date>, d.<COUNT(drug)> FROM (SELECT <pid> FROM [pid]) AS a LEFT JOIN (SELECT <pid>, <sex> FROM [[pid, sex]]) AS b ON a.<pid> b.<pid> LEFT JOIN (SELECT <pid>, COUNT(<drug>) AS <COUNT(drug)> FROM [[pid, drug]] GROUP BY <pid>) AS c ON a.<pid> c.<pid> LEFT JOIN (SELECT na.<pid>, na.<child>, nb.<sex>, nc.<COUNT(drug)> FROM (SELECT <pid> FROM [[pid, child]]) AS na LEFT JOIN (SELECT <pid>, <sex> FROM [[pid, sex]]) AS nb ON na.<child> nb.<pid> LEFT JOIN (SELECT <pid>, COUNT(<drug>) AS <COUNT(drug)> FROM [[pid, drug]] GROUP BY <pid>) AS nc ON na.<child> nc.<pid>) ON a.<pid> na.<pid>; Ex. a28 The above report shows gender and the number of drug prescriptions for individuals as well as the same information for each of the individuals children.

Aggregation is both per child and per individual. The following EVR is also permissible and is created using the minimum cursor principle: [pid, sex, COUNT(drug), child, child.sex, COUNT(child.drug)] Ex. a29 It does not however give the same results as the preceding EVR due to the multiplication effect. Another aspect to pay attention to is that the nested EVR [sex, COUNT (drug) has an implicit PD that is derived from the domain of child. This is a similar concept as the implicit output dimension for nested sets in SDL expression as presented in Ex. Clearly, the EVR notation is much more concise than the -113corresponding SSDL translation, not to mention the SQL translation which can be even longer because of the expansion of the concrete virtual relations.

Pivoting Extended virtual relations can be regarded as an automatic mechanism to represent data as a horizontal structure (Agrawal, et al., "Storage and querying of e-commerce data," in Proceedings of the 27th VLDB Conference, Roma, Italy, 2001). Therefore they are also closely related to pivoting and unpivoting of data, for which SQL extensions have recently been introduced (Graefe, G. and J. Alger, "Electronic database operations for perspective transformations on relational tables using pivot and unpivot columns," U.S. Patent No. 6,298,342 and Johnson, T. and D.

Chatziantoniou, "Extending complex ad-hoc OLAP," In Proceedings of the 1999 ACM CIKM International Conference on Information and Knowledge Management, Kansas City, Missouri, November 2-6, 1999, ACM, 170-179), but are indeed a part of a more general paradigm for representing tables (Gyssens, M. et al., "Tables as a paradigm for querying and restructuring," In Proceedings of the Fifteenth ACM SIGACT-SIGMOD-SIGA R T Symposium on Principles of Database Systems, June 1996, Montreal, Canada, ACM Press, 93-103). Presentation of data in a pivoted form is often desired such as in data mining of association rules or "market basket" analysis (Agrawal, R. and R. Srikant, "Fast algorithms for mining association rules in large databases," in VLDB '94, Proceedings of 20th International Conference on Very Large Data Bases, September 12-15, 1994, Santiago de Chile, J. B. Bocca et al. Eds. Morgan Kaufmann, 487-499).

Next is introduced how the virtual relations can be used to provide such functionality. Instead of introducing new SDL/SSDL keywords, the present invention utilizes new system dimensions. This addition is confined to the construction of VRs and does not impact the semantics of the language or its compilation in any way.

Consider the following virtual relation: sys.dim, value, objid Ex. 114- By definition, this VR should pivot all accessible data in the database into a vertical representation (Agrawal et al. 2001). The PD sys. dim is the "column" with the name of the dimension, value is the dimension value (of data type ANY or STRING), In obj id is a unique id for a tuple. Thus, using SSDL, it should be possible to construct S 5 any other relation from the one in Ex. (a30). How the relation in Ex. (a30) is Sgenerated will depend on the underlying data structures with the present invention multi-table schema, it could become a hefty union statement whereas with a vertical Slayout it would simply be a plain select from a single table. Typically, one is however not interested in the "whole database" and the logic behind the VR should be able to exclude tables that will not provide the right data, as illustrated in Fig. 7. For instance, the following is possible: sys.dim, value.int WHERE sys.dim IN projid 1 AND datatype "integer" Ex. a31 The idea is that any attributes that are available on the dimensions can be used to qualify which dimensions to include in the pivoting, i.e. use a bootstrapping approach where one specifies SDL dimensions using SDL expressions. Similar ideas where the schema or the metadata can be queried to form a data query have been presented in other languages such as SchemaSQL (Lakshmanan, L.V.S. et al., "On efficiently implementing SchemaSQL on a SQL database system, "in Proceedings of the 25th VLDB Conference, Edinburgh, Scotland 1999) and XSQL (Kifer, M., "Querying object-oriented databases," In Proceedings of the 1992 ACM SIGMOD International Conference on Management of Data, ACM Press, 393-402).

To provide a more intuitive and concise notation, it is also possible to let the presence of the dimension value. int imply that datatype "integer".

Furthermore, if a system relation is defined where the PD sys dim is associated with all other dimensions, it possible to imply only pivoting of the concrete relations for which they appear in: sys.dim, value.string, pid WHERE sys.dim IN singleton Ex. a32 -115- The above means pivoting of all dimensions that occur within a relation with the dimension pid that are of type string and are singletons. Note, here the sys. dim is the PD whereas sys. dim. pid is not. The implied constraint is equivalent to sys. dim IN rel. dim "pid" i.e. the dimension has to be in relation with pid. In the same manner, it is possible to introduce a system dimension and corresponding VR that are available for all PD, e.g. [pid, pid. dim. name, pid. dim. value] to access all dimensions associated with pid.

Similarly, dimensions that are collections can be included as well, however, they require the dimension obj id to be meaningful (non ambiguous) in the output.

Consider now the following SDL query: pid I sex "male" AND pid IN SELECT a.<pid> FROM sys.dim, value, pid WHERE sys.dim IN singleton AS a WHERE a.<value> NULL; Ex. a33 The above query finds all males for which all the singleton dimensions in the schema are defined. Although this query has no particular meaning, it demonstrates the power of the data abstraction with virtual relations and the possibility to merge SSDL (SQL) into SDL expressions. Other variants of VRs are also possible and system dimension or dimension attributes that "exist" in multiple relations are for instance useful to constrain the scope of queries, e.g. projects, users etc.

DISCUSSION

In this work applicants have provided SDL, a language for defining sets of objects based on relational data structures. One of the primary goals of the SDL project was to enable non-expert users to issue ad-hoc queries, in particular queries on clinical and genetic data. To achieve this, applicants through the present invention define a language syntax that enables very simple queries to be generated using GUI support such as drag-and drop and relieved the user from logical navigation of relations. While the aim was to make simple queries as easy as possible, applicants have also tried to ensure enough expressiveness such that the invention language covers a sufficiently wide range of query classes. A pragmatic approach was taken and attempts to ensure that the language could be implemented effectively on -116commercial RDBMS were made. As a consequence of this approach, the SDL language can be integrated relatively easily with SQL. This allows any lack of expressive power in the SDL language to be compensated with SQL, either through explicit bi-directional inlining or by augmenting SDL with template functions.

The present invention SDL language provides a simple, yet powerful construct to deal with high-dimensional life-science data as well as event based clinical data. In particular, temporal expressiveness is provided in SDL through the use of binding variables. Therefore, one can say that time is handled explicitly in SDL since the language does not have specific language constructs that deal with time implicitly.

Such implicit handling of time may however be of interest and future research on SDL may involve analysis of the possibility of combining SDL with languages, such as TSQL, designed specifically for temporal databases (Snodgrass, "The temporal query language TQuel," ACM Trans. Database Syst. 12(2):247-298, 1987; Snodgrass, R. Developing Time-Oriented Database Applications in SQL Morgan Kaufman Publishers, Inc., 1999; and Connor et al. 1999). As an example, one can envision extending the implicit join in SDL to include an additional temporal dimension with an implicit overlap predicate. At the moment, such integration is hampered by the unavailability of commercial temporal database systems. It is also possible to envision future additions that are more easily reached. They may include the creation of additional record keywords, e.g. FIRST and LAST, or even additions based on some of the new language constructs in SQL3 (ISO-ANSI. 1997. ISO-ANSI working draft, (SQL/foundation). Tech. Rep. DBL:LGW-008, ISO/IEC JTC1/SC21/ WG3 Database, American National Standards Institute. April 1997) such as order based aggregate operators for relations. Today, these features have to be used through SSDL inlining or template functions.

Although one embodiment is based on a dynamic multi-table RDBMS schema, relational data abstraction is an essential part of the language design. SDL has a cursor free notation, i.e. there is no reference to relation names and cursors do not have to be declared explicitly. With this design approach, expressions with predicates on multiple attributes per tuple (relation) are enforced with automatic record-locking -117or the minimum cursor principle, if there is no ambiguity, and with explicit recordlocking where there are multiple references to the same attribute.

Virtual relations and the CDC concept were also introduced for data abstraction and to enable schema evolution. In an ideal world, where it is possible to design a schema that can capture all expected data, a single RDBMS table could represent each CDC, i.e. each CDC would have a horizontal data structure. Attributes that are missing can be substituted with NULLs. In practice, where the schema evolves and new attributes may be generated, this requires that the RDBMS allows populated tables to be modified new columns are NULL padded for existing tuples), something which is not possible in all commercial systems. Similarly, if a vertical storage model is used, the present invention definition of VR allows data from multiple data imports into several overlapping concrete relations to be automatically NULL padded. Therefore, the definition ofa CDC in SDL allows virtual relations to be generated dynamically based on dimension fingerprints and the registered concrete relations and hides whether those concrete relations are stored in a vertical, horizontal or even federated manner (Haas, L. M. et al, "Data integration through database federation," IBM Systems Journal 41(4) 578-596, 2002 and Haas, L. M. et al., "DiscoveryLink: A system for integrated access to life sciences data sources," IBM Systems Journal 40(2):489-511, 2001).

It should nevertheless be acknowledged that equivalent abstraction could be achieved by assigning a single RDBMS view to each CDC and then alter the definition of the view as the schema evolves. Indeed, one can argue that views, table functions, extendable virtual table interfaces as well as indices are all some form of data abstraction that exists in current database systems. Thus, it is more of a question whether the abstraction is implemented in the "table interface" or the interface between the languages, SDL and SQL, as in one embodiment of the present invention current implementation. The present invention approach is less dependent on the flavor of the underlying RDBMS and is easier and more flexible with regard to implementing dynamic session dependent federation, i.e. where each session may have different concrete relations registered, and where session dependent scoping parameters impact how virtual relations are constructed.

-118- Virtual relations and the emphasis on data abstraction in the SDL system resembles in many ways earlier work on the universal relation model (Maier, et al., "On the foundations of the universal relation model," ACM Trans. Database Syst. 9 283-308, 1984). Although there are similarities, there are also some important differences. In SDL, the present invention introduces the concept of a primary dimension which makes the role of the related dimensions unambiguous and also the concept of implicit joins. Additionally, in SDL the present invention allows pathexpressions with virtual relations that define join-paths unambiguously. Collections and multiple reference to collection dimensions is handled in SDL, however, it is not clear how this can be resolved in the universal relational model without the use of explicit joins and aliases the same thing applies to EVRs in SSDL.

There are some similarities between the construction of virtual relations in SDL and the binding phase in the weak universal relation model. Ideas based on the use of implicational dependencies, as described by Maier et.al. (1984), could possibly be used in the definition of virtual relations. Applicants recognize that with mixed fragmentation of relations (Meghini, C. and C. Thanos, "The complexity of operations on a fragmented relation," ACM Trans. Database Syst. 16(1):56-87, 1991), the current invention definition of virtual relation based on a CDC is inadequate given the specification of the minimum cursor principle. For instance, if there are registered concrete relations for [pid,yob], [pid, yod],and [pid, yob, yod],the SDL system could fail to identify an individual for which data on the same individual is stored in two binary relations. This would happen if the query has an expression with predicates on both yob and yod within the same conjunct. For singleton relations, it is relatively easy to provide a remedy to this problem, either by relaxing the minimum cursor principle or by modifying the construction of the virtual relations.

This type of work-around may still have some performance implications and is not easily extended to collections. For the time being, the present invention does not fully support mixed fragmentation.

Data updates in the universal relational model are non-trivial and require the definition of insert-able tuples (Brosda, V. and G. Vossen, "Update and retrieval in a relational database through a universal schema interface,"ACM Trans. Database Syst.

-119- 13(4):449-485, 1988). Currently, the invention SDL system only supports import into concrete relations that contain the dimension fingerprint of the imported data and therefore complications of this nature do not come up in SDL. Similar issues could however arise if the present invention allows import of data into EVR. Some functionality of this kind might however be very useful such as for populating starschemas. Actually, applicants believe that a functionality that is currently available and of equal or higher importance is the ability to link or register concrete relations with data that has already been imported to the underlying RDBMS. In the invention multi-table layout, dimensions are simply assigned to columns in tables or views.

This makes an external application, or a data administrator, responsible for managing the data and gives full control and direct access to all the indexing features in the RDBMS. In this case, the SDL system provides just read-only access for data analysis.

Related to the previous subject is the fact that the SDL metadata does not enforce integrity constraints although domain definitions are used to validate data that is imported through the SDL import manager. Instead, the SDL system relies on standard RDBMS features for that purpose. The assignment of a domain to dimensions does nevertheless provide implicit logical connections between columns as compared to more explicit connection using foreign-key constraints in the RDBMS.

In principle, foreign-key constraints could be used to provide comparable logical navigation between relations. That approach would not have provided the same flexibility for federation, because such constraints cannot span multiple database instances, and in particular, session dependent federation would be impossible. Also, regular RDBMS foreign-key constraints would not work with vertical data layout.

The metadata in SDL is rather to provide schema abstraction for queries as well as to make the interpretation of data application independent via metadata sharing. In this regard, SDL metadata is quite comparable to XML schemas (Chamberlin 2002), and applicants are convinced that it is possible to devise some mappings between the two since both forms can represent relational data structures (Funderburk, E. et al, "XTABLES: Bridging relational technology and XML," IBM Systems Journal 41(4): 616-641,2002). A key difference between the two is that the P:\OPER\DH~3017l423 Spal doc-15/01/2n00 00 -120same RDBMS relations can be mapped multiple times with different dimensions in SDL, Se.g. relations that store information on multiple objects. Due to its hierarchical form, the XML format has however an inherent one-to-many structure and representation of data t) that has many-to-many or many-to-one connections, such as in DAGs, is not easily ,1 5 supported except with the use of references. Hence, applicants believe that the SDL metadata framework is better suited for data that requires incremental updates and where multiple "views" on the same data tuples is valuable. Such data views are generated in the SSDL system by defining primary dimensions and their related dimensions. For transaction-based data transfer of relatively few objects, XML structure may however be the natural choice. The same applies to reports, especially when objects contain multiple collection attributes. Therefore automatic translation of reports into XML form, based on EVR notation as described in section 6, should be valuable. Furthermore, the possibility of combining SDL metadata with XML schemas as well as the languages SDL and XQuery should be of general interest.

While this invention has been particularly shown and described with references to preferred embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the scope of the invention encompassed by the appended claims.

The reference in this specification to any prior publication (or information derived from it), or to any matter which is known, is not, and should not be taken as an acknowledgment or admission or any form of suggestion that that prior publication (or information derived from it) or known matter forms part of the common general knowledge in the field of endeavour to which this specification relates.

Throughout this specification and the claims which follow, unless the context requires otherwise, the word "comprise", and variations such as "comprises" and "comprising", will be understood to imply the inclusion of a stated integer or step or group of integers or steps but not the exclusion of any other integer or step or group of integers or steps.

Claims (38)

1. A method of retrieving sets of object data from a data store of a computer system t by defining queries on virtual relations which define desired sets of object data to be ¢€3 retrieved from the data store, comprising the steps of: providing a written representation of a desired object data set in terms of dimensions from relations storing object data in the data store; implying constraints on relation instances or dimensions including specifying sets of objects using constraints on dimensions and using a record operator to indicate which constraints apply to a same tuple in a relation, said step of implying constraints enabling length of the written representation to be minimized; and using the written representation, querying the data store and retrieving the desired object data set, including using the record operator to apply multiple atomic constraints on different dimensions in a same tuple in a relation, said querying and retrieving resulting in the desired object data set as query results.

2. A method as claimed in Claim 1 wherein the step of providing a written representation includes using a record operator to express to a user where constraints on different dimensions apply to a same tuple in a relation or multiple constraints on a same dimension apply to multiple tuples.

3. A method as claimed in Claim 1 wherein the step of providing a written representation includes employing any combination of a disjunctive expression and a conjunctive expression; and further comprising the steps of: performing OR-distribution on disjunctive expressions; and eliminating from disjunctive expressions, conjuncts with undefined binding variables.

4. A method as claimed in Claim 1 further comprising the steps of: translating a conjunctive expression including reordering terms of a conjunct; P:OPERXDHUOI142I3 spa Idoc-i 501/200 00 -122- c- moving terms with negation and no binding variables furthest back in the conjunct; placing terms without negation ahead of other terms in the conjuncti and adding a terms with the certain set type of the desired data set to the front of the Sconjunct if all existing terms have negation.

A method as claimed in Claim 1 further comprising the step of automatically enforcing a record-operator where an expression in the written representation without the Srecord-operator is semantically equivalent to the expression with the record-operator.

6. A method as claimed in Claim 1 wherein the data store has a native query engine; and further comprising the step of rewriting the written representation such that upon translation of the rewritten written representation into code for the native query engine, the code is optimized for querying the data store.

7. A method as claimed in Claim 1 wherein the step of providing a written representation includes utilizing a certain symbol to specify hierarchical constraints on dimensions.

8. A method as claimed in Claim 7 wherein the certain symbol is a colon or an equal sign followed by a colon.

9. A method as claimed in Claim 1 wherein the step of providing a written representation includes using an expression with bidirectional inlining of SQL statements and SDL statements; and further comprising the steps of denoting a start of SDL inlining in SQL with a bracket type symbol or referring to SDL metadata in an SQL statement by enclosing it in a bracket-operator.

P: OPERDJHUO[71423 spa I.d¢-I AI01/2DOS 00 -123- A method as claimed in Claim 9 wherein the step of denoting includes denoting the start of SQL inlining in SDL with a square-bracket followed by a certain keyword or denoting the start of SQL inlining in SDL with a curly-bracket followed by a certain Skeyword.

11. A method as claimed in Claim 1 further comprising the step of supporting federated queries by having session dependent federation in which metadata is defined at runtime Sand federation nicknames are generated on demand.

12. Apparatus for retrieving sets of object data from a data store of a computer system by defining queries on virtual relations which define desired sets of object data to be retrieved from the data store, comprising: an input component for providing a written representation of a desired object data set in terms of dimension attributes from relations storing object data in the data store, wherein the written representation specifies sets of objects using predicates on dimension attributes and employing a record operator notation to indicate which predicates apply to a same tuple; and an assembly coupled to receive the written representation, in response the assembly using the written representation to query the data store and retrieve the desired object data set by implying constraints on relations and determining number of dimensions and tuples in a relation to apply predicates, said implying constraints enabling length of the written representation to be minimized, the assembly producing an indication of the retrieved desired object data set as query results.

13. Apparatus as claimed in Claim 12 wherein the written representation includes a record operator to express to a user which predicates on different dimensions apply to a same tuple in a relation and which multiple predicates on a same dimension apply to multiple tuples.

14. Apparatus as claimed in Claim 13 wherein the assembly further reorders terms of a conjunctive expression.

P\OPER .DK30171423 sp doc-I7/OI1/200 00 -124- Apparatus as claimed in Claim 12 wherein the assembly automatically enforces a record-operator where an expression in the written representation without the record-operator is equivalent to the expression with the record-operator. e¢3

16. Apparatus as claimed in Claim 12 wherein the written representation utilizes a 0 certain symbol to specify hierarchical constraints on dimensions.

17. Apparatus as claimed in Claim 16 wherein the certain symbol is a colon or an equal sign followed by a colon.

18. Apparatus as claimed in Claim 12 wherein the input component includes: an editor for composing written representations; and a data browser for enabling user browsing of dimension values and relations of the data store, to assist a user in composing desired written representations, wherein the editor employs a user interface which supports drag and drop of dimensions and relation values in written representations being composed.

19. Apparatus as claimed in Claim 12 wherein the written representation includes an expression with bidirectional inlining of SQL statements and SDL statements.

Apparatus as claimed in Claim 19 wherein said expression denotes SDL portions in an SQL statement with one bracket operator, and denotes SQL portions in an SDL statement with another bracket-type operator.

21. Apparatus as claimed in Claim 12 wherein the assembly supports federated queries of the data store by using prefixes or aliases to map dimension names.

22. Apparatus as claimed in Claim 21 wherein the assembly has session dependent federation in which metadata is defined at runtime and federation nicknames are generated on demand.

P:AOPERDH~UOI71423 lda IdOI m2G8 00 -125- S23. A method as claimed in Claim 1 further comprising the step of: enforcing expressions that have predicates on multiple attributes per conjunct in a non-ambiguous way using automatic record-locking such that the predicates on attributes n from a same relation (CDC) are automatically enforced on the same record.

024. A method as claimed in Claim 23 further comprising the step of: enforcing non-ambiguous expressions with multiple concrete dimensions or virtual Sdimensions or both, using automatic record-locking based on the minimum cursor principle.

A method as claimed in Claim 1 wherein the written representation makes multiple references to a same attribute; and the step of enforcing utilizes explicit record-locking based on the minimum cursor principle.

26. A method as claimed in Claim 1 wherein the step of providing a written representation includes using an expression with extended virtual relations, said extended virtual relations including one of predicates on dimensions and (ii) a WHERE clause within a record operator.

27. A method as claimed in Claim 26 wherein the expression further uses nested record-operators.

28. A method as claimed in Claim 1 wherein the step of providing a written representation includes using an expression with extended virtual relations having dimensions from more than one relation (CDC), and further comprising the step of transparently assembling the virtual relations with equi-joins or outer-joins of primary dimensions, based on the predicates in the extended virtual relation. P:\OPERF~O171423 spal dOC.7/ I200S 00 -126-

29. A method as claimed in Claim 28 wherein the extended virtual relation can contain virtual dimensions; and further comprising the step of transparently generating and assembling the virtual relations based on the minimum cursor principle and path expressions of the corresponding virtual dimensions. r

30. A method as claimed in Claim 29 wherein extended virtual relations are used as 0 relations in SQL. I

31. A method as claimed in Claim 29 wherein the extended virtual relation may include aggregate operators; and further comprising the step of automatically applying SQL GROUP BY conditions and predicates within each CDC relation based on the dimensions listed in the output of the given extended virtual relation or the dimension within a nested record operator.

32. Apparatus as claimed in Claim 12 wherein the written representation has an expression with predicates on multiple attributes per relation; and the assembly enforces the expression with one of automatic record-locking and a minimum cursor principle.

33. Apparatus as claimed in Claim 32 wherein the expression has multiple references to a same attribute; and the assembly enforces the expression using explicit record-locking.

34. Apparatus as claimed in Claim 12 wherein the written representation has an expression with an extended virtual relation, the extended virtual relation including one of predicates on dimensions and (ii) a WHERE clause within a record operator.

35. Apparatus as claimed in Claim 34 wherein the expression further uses nested record-operators. P:OPER\DHUOI71423 palI.doc-17/al/200 00 0 -127-

36. Apparatus as claimed in Claim 12 wherein the written representation has an Sexpression with an extended virtual relation having dimensions from more than one CDC; and tn the assembly performs an equi-join between the CDC's that only have non-null ¢c, 5 dimensions. O O

37. A method of retrieving sets of object data, substantially as described with reference O to the drawings and/or examples.

38. Apparatus for retrieving sets of object data, substantially as described with reference to the drawings and/or examples.