JPWO2005103901A1 - Computer system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve performance and fault tolerance of a computer system. A load table is distributed by holding a copy of a location table or an alternative key location table in an accelerator system, and executing so-called index processing in the accelerator system. This copy is realized with low load by using broadcast communication. A large amount of communication between the accelerator systems 1 to 9 and the primary system 0 is smoothly performed by performing upstream communication by transmission prompting. Improve the performance of the entire computer system. [Selection] Figure 8

Description

  The present invention relates to a computer system and improves the performance of the computer.

  In order to improve the performance of conventional computer systems, the performance of a single CPU has been improved, the main memory has been increased, and a plurality of CPUs have been installed to improve the processing capability. However, the performance improvement of a single CPU has been moderate compared to the increase in processing amount, and rather, the performance improvement by mounting a plurality of CPUs has been the main. In particular, the tendency was remarkable in UNIX (registered trademark) servers. However, contention of main memory, CPU, and external storage device has occurred, and even if the resources are increased, the processing capacity is not proportionally increased. In particular, when the number of CPUs mounted on a computer is about 100, resources required for the control increase, and it is almost impossible to improve the processing capacity. In addition, for the purpose of improving the capacity of the computer, the application computer and the database computer are also separated, but it is necessary to unify the database, and the database has become a bottleneck.

  The present inventor introduced the concept of a location table and an alternative key table in place of the conventional hierarchical index, simplified the complicated processing associated with the index processing, and performed a method such as binary search for searching the table itself. Invented “Data Storage Retrieval System (Patent Documents 1 and 2)” that can secure high speed and easy maintenance. Further, in the “database reorganization system and database (hereinafter referred to as“ database reorganization system ”) (Patent Document 3), the database is operated with respect to the database proposed in the“ data search storage method ”. A mechanism that allows reorganization is proposed, and an alternative key location table is added to the alternative key table to invent efficient reorganization. In addition, in the “database accelerator function (hereinafter referred to as“ accelerator ”) (Patent Document 4)”, the accelerator system holds a copy of the location table and alternative key location table, and records are stored. It has invented that access can be processed in parallel by using the accelerator system location table or the alternate key location table when searching. “Data storage and retrieval system (Patent Document 5)” invents a system in which a connection from a primary block to an overflow block and from an overflow block to an overflow block is performed using an overflow block management table. In the "Data Storage and Retrieval System", the alternate key overflow block management table is also used for the concatenation of the alternate key block and the alternate key overflow block, and the concatenation of the alternate key overflow block and the alternate key overflow block. This is the method used. Furthermore, in the “communication system (Patent Document 6)”, the collision detection type communication which is currently the mainstream of communication has the problem that the collision increases as the communication amount increases, and the communication amount rapidly decreases. A mechanism that does not collide is invented by performing upstream communication triggered by transmission of a transmission reminder message from a rhythm node.

Japanese Patent No. 3345628 U.S. Pat. No. 6,415,375 Patent application PCT / JP03 / 11592 Patent application PCT / JP03 / 13443 Japanese Patent Application No. 2004-020006 Japanese Patent Application No. 2004-094628

  In the computer system, the present inventors have invented an “accelerator” as a means for improving the processing capability. This improves the processing capability of the database processing computer that has been a bottleneck in the past. However, in order to realize an accelerator system, it is necessary to have a copy of the location table and the alternate key location table, but the "Accelerator" only showed how to copy individually. there were. The method is as follows: “In this specification, the change information transmission mechanism, the change information reflection mechanism, and the information transmission / reception are described through the communication path. However, the method can be realized by creating special hardware. Yes, if you change the information at a specific address, the primary system location table and alternate key location table change by automatically rewriting the information at another address in the same way. In such a case, it is a mechanism (synchronous update mechanism) that updates the Fland location table and the Fland alternative key location table of the accelerator system.

Realizing a mechanism to copy the location table and alternate key location table with low load as described here not only improves the performance of the "Accelerator" but also improves the performance of the computer system. It is useful for. In addition, a large amount of communication occurs between the accelerator system and the primary system (the system that processes data records). If this communication is not performed smoothly, the overall processing capacity cannot be improved. End up. Also, how to perform backup efficiently is an important issue in operating a computer system, but it is not only efficient backup and using the backup system as a backup system. Use as a reference system to improve the performance of the entire computer system.

  The present invention has a data record having a data item including a primary key, a primary block for storing the data records in order of the primary key, and a location table entry in which the address of each primary block is described in a continuous area. A primary system that has a location table, an accelerator system that has a location location table entry that contains the address of each primary block in a continuous area, and a communication path When the change information is broadcast from the primary system and the change information is received by the accelerator system, the accelerator system sends the change information to the primary system after a predetermined waiting time. And transmitting the further completion notice, in a computer system.

  The present invention also includes a data record having a data item including a primary key, a primary block for storing the data records in order of the primary key, and a location table entry in which the address of each primary block is described as a continuous area. A primary system having a location table, and an accelerator system having a location table having a location location table entry in which a primary block address is described in a continuous area, The rhythm node sends a transmission dunning message by broadcast, and the accelerator system receives the transmission dunning message. And performing transmission to the stem, in the computer system.

  The present invention also provides a data record having a data item including a primary key and an alternative key, an alternative key entry including the alternative key and the primary key, an alternative key block including the alternative key entry, and an alternative key location A primary system that holds an alternate key location table that has a table entry in a continuous area, and an accelerator that holds an alternate key location table that has an alternate key location table entry in the continuous area. The primary system sends the change information of the alternative key location table, and each accelerator system receives the change information of the alternative key location table, Location table And a change completion notification to the primary system is transmitted from the accelerator system after a predetermined waiting time when the change information of the alternative key location table is received. It is in.

The present invention also provides a data record having a data item including a primary key and an alternative key, an alternative key entry including the alternative key and the primary key, an alternative key block including the alternative key entry, and an alternative key location A primary system that holds an alternate key location table that has a table entry in a continuous area, and an accelerator that holds an alternate key location table that has an alternate key location table entry in the continuous area. System and a communication path, and the rhythm node sends a transmission dunning message by broadcast, and the accelerator system receives the transmission dunning message, and after the predetermined waiting time, from the accelerator system For the primary system And performing transmission, in a computer system.
The present invention also includes a data record having a data item including a primary key, a primary block for storing the data records in order of the primary key, and a location table entry in which the address of each primary block is described as a continuous area. A plurality of accelerator systems having a primary system having a location table and a location table having a location table entry in which the address of each primary block is written in a continuous area. And a communication path and a rhythm node that transmits the transmission reminder message, the rhythm node transmits the transmission reminder message to each accelerator system, and each accelerator system receives the transmission reminder message. Triggered the door, after a predetermined waiting time, and transmitting a processing request for the primary system, in a computer system.

It is an example of the database used as the object of the present invention. This is an example in which an overflow block management table is used in a database that is a subject of the present invention. It is a figure explaining the principle of an accelerator. It is a figure in case there is one accelerator system. It is a figure when an overflow block management table is used in the case of one accelerator system. It is a time chart in the case of performing uplink communication using the transmission prompt message by broadcast communication. This is a typical configuration in a communication system. This is a computer system configuration that uses a plurality of accelerator systems using a communication system. It is a time chart which performs an update process and change completion notification with respect to transmission of change information and a transmission reminder message. It is a figure of the format of the information of a change notification. FIG. 5 is a diagram in which one accelerator system belongs to one group in a group determination method. FIG. 6 is a diagram in which one accelerator system belongs to a plurality of groups in a group determination method. FIG. 11 is a table for confirming that the primary system has received a change completion notification from all applicable accelerator systems. Program logic in the accelerator system that relates to processing requests to the primary system. The program logic in the accelerator system relates to receiving change information from the primary system and performing change processing. Program logic in the primary system that relates to processing requests from the accelerator system. Program logic in the primary system that relates to sending change information to the accelerator system. In the configuration of the computer system, it is a diagram in which functions of a tonic node, a dominant node, and a rhythm node are mounted on a final stage communication device. FIG. 2 is a diagram of a BUS type computer system using an accelerator system. It is a time chart in the case of performing uplink communication using a transmission prompt message by broadcast communication in a BUS type computer system. 1 is a diagram of a ring-type computer system using an accelerator system. FIG.

Explanation of symbols

0 ... Primary system (Tonic node)
01, 02 ... Secondary system 1, 2, 3, 4, 5 ... Accelerator system (dominant node)
6, 7, 8, 9 ... Accelerator system (dominant node)
10 ... Location table 101 ... Final pointer 11 ... Primary block 12 ... Overflow block 15 ... Overflow block management table 151 ... Overflow block management table pointer 15A, 15B , 15C... Alternate key overflow block management table 15A1, 15B1, 15C1... Alternate key overflow block management table pointer 16... Fland location table 161 ..Fland last pointer 16A1, 16B1 , 16C1 ... Fland substitute key table final pointer 19 ... Fland overflow block management tables 19A, 19B, 19C ... Fland substitute key overflow block Management table 19A1, 19B1, 19C1 ... Fland alternative key overflow block pointer 20, 21, 22, 23 ... Rhythm node 41,42 ... Communication path 431,432,433 ... Branch line communication paths 441, 442, 443 ... Branch line communication paths 49 ... Switch 50 ... Intermediate communication devices 55, 56, 57 ... Final communication devices 61, 64, 65, 66 ... Communication channels 641, 642, 643 ... communication paths 71, 74, 75, 76 ... communication paths 741, 742, 743 ... communication paths 110, 120, 130, 140, 150 ... block element device 160, 170, 180, 190... Block element devices 111, 121, 131, 141, 151... Programs 161, 171, 181, 19 ... Program 112, 122, 132, 142, 152 ... Block or alternative key block 162,172,182,192 ... Block or alternative key block 200,210,220 ... Table element device 201, 211, 221 ... program 202, 212, 222 ... location table or alternative key block 300 ... communication processing device S1 ... transmission reminder S2, S3, S4, S5 ... uplink communication S6, S7, S8, S9 ... Downlink communication

[Data storage search method]
Here, the contents of the data storage search method invented by the present inventor will be briefly described. The data storage / retrieval method of the present invention uses a location table and an alternative key table and performs a binary search on them to search for a target record. Data records are stored in the primary block in the order of primary keys. When adding a data record to the primary block, if the primary block is full, the overflow block is concatenated to the primary block to store the data record. It is possible to further connect an overflow block to the overflow block. It has a location table having a location table record (or location table entry) in which the address of each primary block is described in a continuous area. The location table is secured in advance in a continuous area. Here, the continuous area is a logical order, and may be separated in a physical area. In such a case, it can be treated as logically continuous using the address conversion table. The same applies to the following description. The last pointer is used to indicate the end of the used area of the location table.

  Records are stored in a fixed-length storage area called a block. A block consists of a primary block and an overflow block. If a record cannot be added to the last primary block, a primary block is added after that and the record is stored.

  Here, concatenation does not mean that the physical block is physically connected, but the primary block holds the address of the first overflow block, and the first overflow block is the second overflow block. Since the state in which the block address is held can be handled as if the block is physically connected, such an expression is used (the same applies hereinafter). Because of this storage, the location table entries are arranged in the primary key order. A search by primary key finds a block by performing a binary search between the first address in the location table and the location table entry pointed to by the last pointer, and the target record in that block. Find out. If an overflow block is connected to the block, the overflow block is also searched. Here, the search is described, but the update, addition, and deletion of records can be realized with the same logic.

  An alternative key record (or an alternative key entry) composed of an alternative key value and a primary key value is stored in the alternative key block in the order of the alternative key values. When adding an alternate key entry to the alternate key block, if the alternate key block is full, concatenate the alternate key overflow block to the alternate key block and store the alternate key entry To do. It is possible to further concatenate an alternative key overflow block to the alternative key overflow block. It has an alternative key location table having an alternative key location table record (or an alternative key location table entry) in which the address of each alternative key block is described in a continuous area. The substitute key location table is secured in a continuous area in advance. The alternate key end pointer is used to indicate the end of the used area of the alternate key location table. When an alternate key entry is added and an alternate key entry with an alternate key value greater than the alternate key value of an existing alternate key entry is stored in the last alternate key block and cannot be stored in that alternate key block A new substitute key block is created and the record is stored in the substitute key block. The alternative key location table and the alternative key block are referred to as an alternative key table.

  The substitute key is a non-unique key in the database, such as a name and date of birth in the employee database. There may be no substitute key for a certain type of database, and there may be a plurality of substitute keys. A method for searching for a record with a certain alternate key is to perform a binary search between the first entry in the alternate key location table and the alternate key location table entry pointed to by the alternate key end pointer. Find the desired alternate key block and search within the alternate key block to find an alternate key entry with the desired alternate key. If an alternative key overflow block is concatenated to the alternative key block, the alternative key overflow block is also searched. Next, the primary key of the alternate key entry is used to perform a binary search of the location table, find the target block, and find the target record from within the block. If an overflow block is connected to the block, the overflow block is also searched. Since the substitute key is non-unique, there may be a plurality of records having the same substitute key value. In this case, if the next substitute key record in the substitute key block has the same substitute key value, the same operation as described above is repeated. Here, the search is described, but the update, addition, and deletion of records can be realized with the same logic. When there are a plurality of alternative keys, the same number of alternative key tables as the types of the alternative keys are created and used. This completes the description of the “data storage search method”.

[record]
A record always has one unique primary key and zero or one or more non-unique keys (alternative keys. As a result, there is no problem even if it is unique). In addition to this, it is possible to have items (columns) that are not keys. For example, in the case of an employee database, the primary key is a code that can identify the employee, such as an employee code, and the alternative key is a name, date of birth, etc. It doesn't matter. In addition, for a record in which there is no item that should be a meaningful primary key, a sequential number in the storage order may be assigned and used as the primary key. An item (column) is a unit of information, and there are a key and a key. There is one or more in the record. The column may be in a fixed length format, but may be in a variable length format. In the case of the variable length format, a column for which no column information exists can be recognized as a column.

  In a database system, expressions such as a subschema and a schema are generally used. However, in the present specification, description will be made regardless of any particular case. In this description, expressions such as “add column to database” and “access to database” represent operations on a particular type of database file (for example, employee files) and are stored in the database system. It is not for the entire database file being used. When a specific type of database file is composed of a plurality of database files, the expression database is also used for each database file.

  FIG. 1 shows the structure of the invented database in the “data storage search method”. Here, the overflow block and alternate key overflow block are omitted. FIG. 2 illustrates a mechanism for managing an overflow block and an alternative key overflow block using the overflow block management table for the database of FIG. 1 in the “data storage and retrieval system”. .

[Principle of accelerator system]
An accelerator is literally an accelerator. In the conventional database system, the method for improving the processing capacity of the database is extremely limited. On the other hand, it is possible to improve the processing capacity of the database virtually indefinitely. The principle of the accelerator system is as follows. A binary search is performed on the location table and the alternate key location table to find a primary block including a target record and an overflow block concatenated therewith. Furthermore, a record is found from among them. When a binary search is performed on the location table, the bisection point is searched many times, and this number of times is generally larger than the number of times of searching for a record in the block. Also, the possibility of requesting records in the same block from a plurality of processes at the same time is considerably low. Therefore, if a plurality of copies of the location table and the alternative key location table are held and a binary search can be performed on each of them in parallel, it is possible to execute a large number of access requests for records. FIG. 3 illustrates the principle of the “accelerator”. FIG. 4 illustrates the case of a method that does not use the overflow block management table regarding the basic configuration of the accelerator system. FIG. 5 illustrates the case of the method using the overflow block management table regarding the basic configuration of the accelerator system.

  The primary system is a system that has a primary block and an overflow block, an alternative key block and an alternative key overflow block, and adds, updates, and deletes records. On the other hand, the accelerator system has a Fland location table and a Fland alternate key location table, and performs a binary search on them. After finding the desired primary block and its overflow block, the alternate key block and its alternate key overflow block, on the primary system, the primary block and its overflow block, Operate on an alternate key block and its associated alternate key overflow block.

  FIG. 4 shows a case where there is one accelerator system. The accelerator system has a location table (Fland location table) and alternate key location table (Fland alternate key location table), but the primary block, overflow block, alternate key block, Does not hold alternate key overflow block. The accelerator system's Fland location table is functionally equivalent to the primary system's location table. Similarly, the accelerator system alternate key location table of the accelerator system is functionally equivalent to the alternate key location table of the primary system. Each record in the accelerator system's rand location table points to the same block as each location table record in the primary system.

[Access by accelerator system]
In the accelerator system, when there is an access request, in the case of a primary key, a binary search is performed on the land location table, a target block is searched, and a record search in the block is requested to the primary system. In the case of an alternate key, perform a binary search of the franc, alternate key location table, find the desired block, find the desired alternate key record from the alternate key block held by the primary system, and replace it. Use key records to perform a binary search of the Fland location table to find the desired record. Although the search method has been described here, it is possible to update, add, and delete records by applying this method. Also, in the case of an alternate key, a binary search of the Fland location table is performed based on the alternate key record, but if the alternate key record holds the block address or block number It is unnecessary. In this way, it is possible to increase the amount of processing by searching and updating records in parallel with a plurality of accelerator systems.

  In FIG. 4, the accelerator system has one location table and three alternative key location tables, which is the same number as the primary system. This is called a symmetric system in "Accelerator". On the other hand, for example, an accelerator system that holds only two types of location table and alternative key / location table is assumed, and an accelerator such as only the location table and only the alternative key / location table is assumed. It is possible to create a translator system. This is called an asymmetric system. With respect to the accelerator system, the primary block and the overflow block, the alternative key block and the alternative key overflow block can be handled in substantially the same manner.

  FIG. 5 shows an example in which an overflow block management table is added to the accelerator system described above.

[Information synchronization between the primary system and accelerator system]
Next, a method for synchronizing the information of the primary system and the accelerator system in a system using the “accelerator” will be described with reference to FIGS. 4 and 5. The primary system has a location table 10 and alternative key location tables 10A, 10B, and 10C. Furthermore, it has final pointers (101, 10A1, 10B1, 10C1). When the database has a format using an overflow block management table (in the case of FIG. 5), the database has an overflow block management table 15 and alternative key overflow block management tables 15A, 15B, and 15C. Further, the overflow block management table is provided with an overflow block management table pointer 151, and an alternative key overflow block management table pointer 15A1 is provided for the alternative key overflow block management table 15A. Alternate key overflow block management table pointers 15B1 and 15C1 are provided.

  The accelerator system 3 has a land location table 16, a land alternate key location table (16A, 16B, 16C), and a final pointer (161, 16A1, 16B1, 16C1). When the database is in a database format using an overflow block management table, a land overflow block management table 19 and a land alternative key overflow block management table 19A, 19B, 19C are provided. For each of the alternative key / overflow / block management tables 19A, 19B, and 19C, an alternative key / overflow / block management table pointer 19A1, 19B1, and 19C1 is provided.

  In the accelerator system, when a change occurs in the location table or the alternate key location table in the primary system, the accelerator system is notified of the change, and the accelerator system・ Change location table or alternate key for location ・ Location table. In the primary system, a change occurs in any of the location table 10, the final pointer 101, the alternative key location table (10A, 10B, 10C) and the final pointer (10A1, 10B1, 10C1) of the alternative key location table. If this happens, notify the accelerator system of the change. In the accelerator system, based on the notification, the Fland location table 16, the Fland alternate key location table (16A, 16B, 16C), and the last pointer of the Fland alternate key location table (16A1, 16B1, 16C1) The change portion is changed for any of the above.

  If the database uses an overflow block management table, in addition to the above, in the primary system, the overflow block management table 15, the overflow block management table pointer 151, the alternate key overflow block management If a change occurs in the table (15A, 15B, 15C), alternate key overflow, block management table, and final pointer (15A1, 15B1, 15C1), the accelerator system is notified of the change, and the accelerator In the system, the relevant overflow block management table 19, the final pointer 161, the full overflow block management table pointer 191 and the alternate key over Row block management table (19A, 19B, 19C), for any Furando alternate-key overflow block management table final pointer (19A1,19B1,19C1), and changes the corresponding portion.

  In this way, the change part is notified from the primary system to the accelerator system, and the change is immediately applied by the accelerator system, so that the accelerator location table 16 and the rand overflow block of the accelerator system are applied. Management table 19, Fland final pointer 161, Fland overflow / block management table / final pointer 191, Fland alternative key / location table (16A, 16B, 16C), Fland alternative key / location table final pointer (16A1, 16B1, 16C1) ), Alternate key overflow block management table (19A, 19B, 19C), alternate key overflow block management table, last poi Ter (19A1,19B1,19C1) is kept equal to the primary system. In the accelerator system, when the change of the location is completed, a change completion notification is transmitted to the primary system. The primary system waits for exclusion of this part until the change completion notifications from all accelerator systems arrive.

  The application to the accelerator system in the basic case has been described above. However, in the case of direct column addition, direct column deletion, and direct column change in the “database system”, the following conditions are further satisfied: Join. In the primary system, in addition to the above conditions, the column operation pointer for the working location table, the column operation completion pointer, the new location table, and the new column operation pointer are increased. To cope with this, the accelerator system adds a column operation pointer (Fland column operation pointer), a Fland column operation completion pointer, a new Fland new location table, and a new Fland new column operation pointer for the current working location table. . If the database is formatted using an overflow block management table, a new overflow block management table and a new overflow block management table pointer are added to the new location table. When there is a change in the above elements in the primary system, the change is notified to the accelerator system, and the accelerator system changes the location. In addition, when reorganization is performed, a new location table, a new alternative key location table, and final pointers for each are further added.

  Access on the accelerator system only changes the access to the block to the primary system, and the others are by combining the explanations in the column addition / deletion / change described above and the method explained in the accelerator system. realizable.

  In the above, explanation was made assuming a symmetric system of accelerator system. However, in the case of an asymmetric system, only the accelerator system that holds the changed part in the primary system receives and updates the change. Will do. The above has been described using “quotes from the“ database system ”by the present inventor.

[Communications system]
Next, the “communication system” will be briefly described. The mainstream of current communication is the collision detection type. This is because nodes connected to the communication path (generally including communication devices, but for computers such as personal computers and servers, excluding communication devices) send information at their own discretion, If two or more nodes transmit at almost the same time, a collision is detected, the information is canceled, and transmission from the node is performed again. In the retransmission, the timing is shifted so that a collision does not occur. However, when the amount of communication on the communication path increases, the collision is accelerated.

  On the other hand, in the “communication system”, nodes are classified into tonic nodes and dominant nodes, and host computers and servers are basically tonic nodes. Also, the direction of communication is separated from the tonic node to the dominant node (downlink communication) and from the dominant node to the tonic node (uplink communication). In addition, in order to avoid a collision of upstream communication, a transmission prompt message is transmitted from the rhythm node to each dominant node, and each dominant node passes a predetermined waiting time after receiving the transmission prompt. Later, upstream communication is performed. This predetermined waiting time is determined quantitatively according to the speed of the communication channel, the distance of the communication channel, and the length of the communication message according to the communication sequence number of each dominant node. The node with the communication sequence number n transmits at the same time as the transmission of the node with the communication sequence number n-1 is completed (strictly, the timing differs depending on the length of the communication path). The formula for calculating the predetermined waiting time is described in detail in the “communication system”, and is as follows.

  The transmission message length is L (bytes), and the transmission speed of the communication path is S (bits / second). In contrast, the waiting time is T (nanoseconds). The length of the communication path (62, 63, 64, 72, 73, 74) is as follows: communication path length (62) = communication path length (72), communication path length (63) = communication path length It is assumed that the length (73), the length of the communication path (64) = the length of the communication path (74). Further, it is assumed that the length of the communication path (62) = the length of the communication path (63) = the length of the communication path (64). First, consider a case where the lengths of the communication paths (621, 622, 623) are the same. Each length is D (meters). Further, it is assumed that the length of the communication path (62) = the length of the communication path (63) = the length of the communication path (64). Assume that upstream communication is performed in the order of accelerator systems 1, 2, and 3. Then, at the moment (T1) when the accelerator system 1 finishes transmitting, the end of the transmission information passes through the portion where the communication path 621 is connected to the accelerator system 1. If the transmission message is transmitted from the accelerator system 2 at the time T1, the transmission message from the accelerator system 1 will not collide. The waiting time Tn of the accelerator system n is expressed by the following formula 1.

  Other cases where the lengths of the communication paths are different are described in the “communication system”.

  In downlink communication, a communication right is assigned from the rhythm node to the tonic node, and downlink communication is performed from the tonic node having the communication right. Furthermore, if the communication path is separated for uplink and downlink, more efficient communication can be performed.

  FIG. 6 shows a time chart of transmission prompting and uplink communication when the communication path lengths are equal. In this figure, the rhythm node is built in the tonic node. The number of dominant nodes is four. The distance between the tonic node and each dominant node shall be equal. A transmission reminder is transmitted from rhythm node 0 (tonic node) (S1). Since the lengths of the communication paths are equal, the transmission reminder arrives simultaneously at each dominant node (1, 2, 3, 4). The dominant node 1 immediately performs uplink communication. The dominant nodes 2, 3, and 4 perform uplink communication after a predetermined waiting time. When one cycle of uplink communication is completed, downlink communication is performed from the tonic node. In this figure, it is a figure up to the stream communication to the dominant node 3.

[Means for realizing high-performance computers]
By using an accelerator system, a high-performance computer system can be realized. In conventional computer systems, high performance is achieved by installing a large amount of CPU and main memory. However, as described in the background art, even if a large number of them are mounted, it is difficult to increase the processing capacity because the load for controlling them increases. By using a computer system equipped with an accelerator system, the processing performance of the computer can be dramatically increased. The primary system and the accelerator system may be housed in a single casing, or each may be independent and connected. However, in order to exhibit high performance, it is not sufficient to simply fit in a single housing or connect.

  In “Accelerator”, as a method of notifying the change information of the primary system, a method of sending information individually from the primary system to each accelerator system was shown. Although the explanation is such that information is transmitted / received through the information transmission mechanism, change information reflection mechanism, and communication mechanism, it can also be realized by creating special hardware. If the primary system location table and alternate key location table are changed, the address of the accelerator system is・ Update location table and alternate key location table It is so that a mechanism (synchronous updating mechanism). "That only has been described, the specific implementation means was not clear. Furthermore, although the system is such that a large amount of communication from each accelerator system to the primary system occurs, the means for smoothly performing this communication has not been clear.

  FIG. 7 shows a typical configuration of the “communication system”. Node 0 (0) is a tonic node and has a built-in rhythm node 20. Nodes 1, 2, 3,..., 9 are dominant nodes. Both are connected by a communication path via the intermediate communication device 50 and final-stage communication devices 55, 56, and 57. The communication path is separated for uplink communication and downlink communication. FIG. 8 shows this structure applied to the “accelerator”. The tonic node becomes the primary system 0 and the dominant node becomes the accelerator system (1, 2, 3,..., 9).

[Send Change Information]
A description will be given of how the “accelerator” is realized when such a configuration is adopted with reference to FIG. In this figure, the communication path (61, 64, 65, 66, 641, etc., 71, 74, 75, 76, 741 etc.) and the communication device may be serial transmission or parallel transmission. Further, in FIG. 8, two communication paths are used for uplink communication and downlink communication, for example, 61 and 71 in FIG. 8, but the communication paths 61 and 71 are used as one communication path. There is also a method for realizing uplink and downlink communications by time division. The time chart of FIG. 9 shows such time-division communication. When two communication paths are used for uplink and downlink, uplink and downlink communications can be performed simultaneously in parallel. In the following description, communication from the primary system to the accelerator system is referred to as downlink communication, and communication from the accelerator system to the primary system is referred to as uplink communication. First, the types of uplink communication and downlink communication will be described. When there is some change in the primary system and the change is related to the accelerator system, the primary system transmits the information to the accelerator system as downlink communication (transmission of change information). In response to this change information, each accelerator system applies the change and, when the change is completed, transmits a change completion notification to the primary system as upstream communication (transmission completion notification transmission). In addition, the accelerator system searches the location table and the alternative key location table based on the request from the application computer, and searches for the block and alternative key block based on the search result. Request the system (send processing request). In the primary system, based on a request from the accelerator system, the contents of the block and the alternative key block and the notification of the change result are transmitted as downlink communication (transmission of processing result).

  Here, a method for efficiently transmitting change information will be described. When there is one accelerator system or several accelerator systems, it is unlikely that special problems will occur even if the change information is sent to each accelerator system individually. If this number increases, the time required for transmitting the change information increases, and there is a possibility that other downlink communications cannot be performed. In order to avoid such a situation, the change information is transmitted to the accelerator system by broadcast. It is very efficient to use broadcast because changes in the primary system need to be sent to all accelerator systems in the case of a symmetric system. FIG. 9 is a time chart when the broadcast is used for transmitting the change information in this way. Again, the rhythm node is built in the primary system. After the change information is transmitted by broadcast from the primary system, a transmission prompt message is transmitted from the rhythm node to the accelerator system (1, 2, 3, 4) by broadcast (S1). The change information can also be used as a transmission prompt message. In this case, the relationship between the rhythm node and the primary system is that the rhythm node gives the primary system the right to transmit change information when one cycle of uplink communication is completed. It is necessary to separate from normal downlink communication. Each accelerator system performs update processing immediately after receiving the change information. Thereafter, the accelerator systems 2, 3, and 4 send an update completion notification to the primary system after taking a predetermined waiting time (S2, S3, S4, and S5). FIG. 9 shows the timing when each accelerator system has equivalent performance (the time required for the update process is equal).

[Contents of change information]
The details of the change information will be described in detail. FIG. 10 is an example of change information. The contents of the change information include information such as a destination address 80, a transmission source address 81, a serial number 82, a change target 83, a change target address 84, a change length 85, and a changed content 86. The destination address 80 is an address of a node that should receive the information. In this case, since the change information is broadcast, the accelerator system group ID becomes the destination address. The source address 81 is the primary system address. The serial number 82 is used to distinguish different information when a plurality of change notifications are transmitted and the change completion notification is received by the primary system. The change completion notification is sent to the primary system from the accelerator system in the serial number of the change completion notification to identify the change completion notification for which change information, and is generally used in information communication Is the method. Whether the change object 83 is a location table or a plurality of alternative key location tables, the last accelerator, an overflow block management table, a plurality of alternative key overflow block management tables, etc. -It shows which component is the change among the components that make up the system. This ensures that the primary and accelerator systems have the same correspondence table and can be changed without fail. The change target address 84 can be physically designated or logically designated.

  The term “physical” is used when the components of the accelerator system in the primary system and the components in the accelerator system have the same physical configuration. The physical configuration is exactly the same. For example, the structure of the location table is exactly the same in the primary system and the accelerator system, and on the storage device of the location table in the primary system. The address on the storage device of the location table in the accelerator system is exactly the same. As a specific example, when the location table entry 5 of the primary system is, for example, 50 bytes from the address 1050 of the storage device, the location table entry 5 of all the accelerator systems is That is, 50 bytes from the address 1050 on the storage device. In this case, the primary system is a change notification that the information from which address on the storage device has been changed. When receiving the change notification, the accelerator system changes the change target address of the storage device. In this case, even if there is no information of the change target 83, it operates normally.

The logical is a method of indicating to which entry of the location table, the alternate key location table, the overflow block management table, and the alternate key overflow block management table by an entry number. The other components are not structured to have a plurality of entries. For example, if the final pointer is specified, the target is determined by that.
If this logically the same is further expanded, the structure of the location table and the land location table, and the alternate key location table and the land alternate key location table may be different. The difference in structure is as follows. The information required for the location table is the entry number and the address of the primary block that the entry points to. In addition, in the location table of the primary system, it is effective to expand the entry information so as to hold the statistical information of each block. On the other hand, the accelerator system's Fland location table simply holds the entry number and the primary block address without having statistical information. This saves the amount of information in the accelerator system and reduces changes.
Further, the accelerator can be extended as follows. The accelerator system is supposed to maintain a location table and an alternate key location table. However, when the load of some specific substitute keys is high, the substitute key block is held by an accelerator. Naturally, the substitute key block has the same content as that held by the primary system, and is synchronized with the primary system by the synchronous tight coupling method shown in the “backup recovery system”. By doing so, it is possible to reduce the load on the primary system when the load on some of the alternative key location tables is high.

  The shorter the change information, the shorter the time required for transmission and update. Therefore, it is preferable if the change information can be shortened. For example, in the case of an update to a location table entry, the block number and address are not changed, so the case where the minimum value or the maximum value of the primary key value is changed is targeted. Even when either the minimum value or the maximum value is changed, if the entire primary key value is changed, the lower digits may be changed. In such a case, the change information can be shortened by transmitting only the difference using the change target address and the length 85 to be changed. In this case, when the change target address is logical, it is necessary to use an entry number and a displacement within the entry. Even when the difference is not used, the length to be changed is indispensable when the change target address 84 is physically designated. Even when the change target address 84 is logically specified and not the transmission of only the difference, it is useful for ensuring the change to transmit the length to be changed.

  The content after change is the content itself to be changed. For example, when the maximum value of the primary key value of the location table entry is changed from “56789” to “56790”, “56790” becomes the changed content. Further, as described above, it is also possible to transmit only the last two digits “90” as a difference.

[Broadcast communication]
The communication information is assumed to be performed using broadcast communication, which will be described in more detail. In the case of broadcast communication, there are simultaneous broadcast and group broadcast. Accelerator systems include symmetric systems and asymmetric systems. The symmetric system is a form in which all of the accelerator systems have the location table and the alternate key location table that are held by the primary system. On the other hand, the asymmetric system is a form in which a specific accelerator system holds only a part of the location table and the alternative key location table held by the primary system. The asymmetric system is effective in saving the storage area when the access frequency differs for each key type.

  If all accelerator systems are symmetric systems, simultaneous broadcast is acceptable. On the other hand, in the case of an asymmetric system, group broadcast is used. There are various methods for defining this group, but the method shown in FIG. 11 is representative. FIG. 11 shows a method of allocating one group to one accelerator system by patterning whether the accelerator system has a location table or an alternative key location table. Such a group determination is a method generally performed in conventional communication. This method has the following characteristics. Since each accelerator system has a group ID, the correspondence is easy to understand. On the other hand, when the types of alternative keys increase and the combinations become complicated, the types of groups may increase geometrically. On the primary system, it is necessary to analyze which group is affected when a change is made to the location table or the alternate key location table. Further, in the case of FIG. 11, when the location table is changed, broadcast must be performed for each of group 1, group 2, group 3, and group 4, and the amount of information for downlink communication is reduced. It will increase.

  On the other hand, FIG. 12 is an example in which the accelerator system is allowed to belong to a plurality of groups. Reference numeral 97 in FIG. 12 indicates whether each accelerator system has a location table or an alternative key location table. On the other hand, reference numeral 98 in FIG. 12 shows a method for determining a group name depending on whether a location table or an alternative key location table is held. Reference numeral 99 in FIG. 12 shows a list of group IDs determined in this way. The accelerator system 1 belongs to any of groups 1, 2, 3, and 4. Similarly, the accelerator systems 2, 3, 4 belong to the groups 1, 2, 3, 4. The accelerator system 5 belongs to the groups 1, 2, and 3, but does not belong to the group 4. The accelerator system 6 belongs to the groups 1 and 2 but does not belong to the groups 3 and 4. The accelerator system 7 belongs to group 1 only. When the method shown in FIG. 12 is adopted, a change to the location table or the alternative key location table may be broadcast to a group corresponding to the location table or the alternative key location table. Downlink information communication may be performed once. In addition, the destination of downlink information communication from the primary system can be easily determined. As the alternative key type increases, the group type increases only proportionally. On the other hand, in the accelerator system, it is necessary to hold which group it belongs to so that it can cope with broadcast communication for a plurality of group IDs.

  In FIG. 9, after receiving the change completion notification, downlink communication is performed from the primary system (S6, S7, S8, S9). In practice, however, the communication paths for uplink communication and downlink communication are separated. Since it is provided, downlink communication may be performed when transmission of the change information and transmission of the transmission reminder are completed. Further, downlink communication does not need to be performed equally for each accelerator system, and may be performed according to the amount of communication.

[If the performance of the accelerator system is uneven]
When the performance of each accelerator system is not equivalent, the waiting time is measured after receiving the change information and the transmission reminder. The update processing time can be made a waiting time by transmitting the time required for the primary system in the change information and adding the updating processing time to a predetermined waiting time. In this case, the performance of the accelerator system 1 (communication sequence number is the first accelerator system) should be the same as that of the primary system, and the accelerator system with the communication sequence number 2 or later is predetermined. It is necessary to have enough performance to finish the update process within the waiting time. If the update process is not completed within a predetermined waiting time for some reason, the change completion notification is not transmitted. This is because if a change completion notification is transmitted at an arbitrary timing, a communication collision occurs.

[Change completion notification]
Now, broadcast communication is performed for the group in this way, the necessary changes are made in the accelerator system, and a change completion notification is sent from the accelerator system to the primary system. From the viewpoint of the primary system, it is necessary to receive and confirm many change completion notifications without omission. As a method for confirming without omission, it is effective to use a change completion confirmation table shown in FIG. Each node in the change completion confirmation table is an accelerator system. The status is incomplete or complete. In FIG. 13, incomplete is indicated by a blank. The top row count is the number of nodes in the group first. Each time the primary system receives a change completion notification from each node, the status of that node is set to completion, and 1 is counted from the count. Decrease. If a change completion notification comes again from a node whose status has already been completed, it is preferable to make an inquiry to that node. In this case, 1 is not subtracted from the count. If the count reaches 0, it can be confirmed that all the change completion notifications for the group have been received. It is convenient to create a copy of this change completion confirmation table for each group and make a copy of the copy when sending a change notification from the primary system.

  The primary system sends a change notification to the accelerator system, and until the primary system receives a change completion notification from all target accelerator systems (accelerator systems that belong to the change notification target group) -In the system, the changed part is excluded.

  In the above, the transmission of the change information and the transmission of the change completion notification have been described. Next, a method for transmitting a processing request from an accelerator system is to transmit from each accelerator system using a transmission prompt indicated by “communication system”. In this case, there is a high possibility that the length of the previous change completion notification and the information of the processing request is different. This is because the change completion notification only needs to include the serial number of the change information and information indicating completion, so that a length of about 20 bytes is sufficient even if the destination address and the source address are included. On the other hand, in the case of a search, a processing request requires a block address, a key type, a key value, and the like. Updating and writing require the contents of the record, which can be even longer.

  In this way, if information having different lengths is handled with a single-length packet, useless space in the packet becomes large. In order to avoid such a state, as described in the “communication system”, it is preferable to specify the type of transmission information from the accelerator system when transmitting a transmission reminder. In the transmission reminder in the case of a change completion notification, the packet length is, for example, 64 bytes, and in the case of a normal processing request, the packet length is, for example, 256 bytes. Of course, if the packet length is different, the required waiting time will vary for the same accelerator system, so if you use multiple packet lengths, you need to wait for a certain amount of time to accommodate them. It is necessary to calculate time.

  As described above, by applying the “communication system” to the “accelerator” and transmitting the change information by broadcast, efficient downlink communication can be performed.

[Send processing request]
The uplink communication processing request has already been briefly described, but will be described again. Each accelerator system accesses a database based on an access request to the database from an application program in the accelerator system or an application processing computer. For example, when the access is based on the primary key, the accelerator system searches its own location table and finds a target block. Blocks and overflow blocks do not exist on the accelerator system, only on the primary system. In the case of a search request, a search DB type, a search target block type, a block number or block address of a target block, and a primary key value are designated. It is sent to the primary system as a processing request record. Since this processing request is generated at an arbitrary timing when each accelerator system is operating, there is a possibility of collision if they are transmitted without permission. In addition, since collisions are likely to occur when communication is congested, performance deteriorates in a state where many processes must be performed.

  In order to avoid such a collision, in the “communication system”, uplink communication is performed by transmitting a transmission reminder from a rhythm node, and when the communication reminder is received by a dominant node (accelerator system), after a predetermined waiting time. Uplink communication is to be performed. By doing so, a collision does not occur in uplink communication. The method of transmitting the transmission reminder and performing the upstream communication with a predetermined waiting time after receiving the transmission reminder is also applied to the “accelerator”. The timing chart in this case is the same as that in FIG.

[Accelerator system logic]
The accelerator system requires a program, but the logic of the program will be explained again. 14 and 15 are flowcharts of the accelerator system program (accelerator system program). In FIG. 14, first, access is accepted from an application program. The application program may be run in the accelerator system, or may be run in an external system to accept database access. The accelerator system program then identifies the type of access. Which database (ski) is targeted, what is the key type, whether access is search, add, update, delete, what is the key value. Next, a binary search of the location table or alternative key location table is performed according to the key type and key value. There, the number or address of the target block or alternate key block is found. Therefore, the accelerator system program creates a transmission message and stores it once. After that, when a transmission reminder message is received, after a predetermined waiting time, a processing request is made to the primary system by upstream communication. Processing is performed on the primary system, and the result is returned to the accelerator system. The result is returned to the application program. FIG. 15 is a flowchart in the case of receiving change information from the primary system. Receive change information from the primary system. Make the necessary changes according to the change target. Thereafter, a change completion notification is transmitted after a predetermined waiting time.

[Logic on the primary system]
Next, processing contents of the primary system program (primary system program) will be described. 16 and 17 are the flow charts. FIG. 16 relates to a processing request from the accelerator system. First, accept processing from the accelerator system. Next, the type of access is identified. Which database (ski) is targeted, what key type is, whether access is search, add, update, delete, what is the key value, block or alternate key block number. Next, find an entry in the location table or alternate key location table by block or alternate key block number. If the processing request is an update system, the entry is excluded. The exclusion method will be described later. When the processing request is by the primary key, the target record is detected from the primary block and the overflow block connected to the primary block, and if the processing request is a search, the record Set. If the processing request is an update, the record is updated. In the case of addition, it is assumed that there is no record having the primary key value, the record insertion location is determined, and the record is inserted. If the processing request is deletion, the record is deleted. If the processing request is search, update, or delete and there is no record with that primary key value, an error is assumed. If a process request is added and a record with that primary key value already exists, an error is assumed. If it is normal, necessary information such as a record is set in addition to the return code. Then pass the information to the accelerator system. This is transmitted as downlink communication. FIG. 17 is a flow chart for transmitting change information for the accelerator system.

[Exclusion method]
Next, the exclusion method will be described. The simplest way to do this is by setting a flag for the desired entry in the location table or alternate key location table. In this way, it is possible to reliably detect which block or alternative key block is exclusive. As a method of releasing exclusion related to a transaction by transaction termination, deadlock or transaction cancellation, or abnormal termination of the program, the location table entry to be excluded and the alternate key location table entry for each transaction Is written to a separate table and referenced.

[Special accelerator system]
Although the basic description of the accelerator system has been given above, a modification will be described. If the number of accelerator systems is large, the primary system accepts the processing request from the accelerator system and processes it. If a change occurs, the change information is sent to the accelerator system. It is useful to improve the overall efficiency to perform only the function. However, when the number of accelerator systems is not so large, the overall processing efficiency can be improved by placing the functions performed by the accelerator system on the primary system. In such a case, for example, the accelerator system 1 in FIG. 8 is a dummy, and the accelerator system 1 does not perform any processing on the processing request received from the application program by the accelerator system 1. It is a method of transmitting to the primary system as it is. The accelerator systems 2, 3, 4,... 9 perform processing as shown in FIG. In this case, the primary system program has the logic shown in FIG. 14 and executes the logic of FIG. 14 and the logic of FIG. 16 in the case of a processing request from the accelerator system 1. .

[When there are many accelerator systems]
The example in the case of adopting the basic configuration has been described above. In the case where the number of accelerator systems is relatively small, it is considered that no particular problem occurs even if the transmission reminder message is transmitted to all accelerator systems. On a communication path having a speed of 10 Gbps, when 8 bits are set to 1 bit, communication of 10 bits per 1 ns (nanosecond) is possible, and 0.8 ns is required for 1-byte communication. The length of the uplink communication message from the accelerator system is constant, for example, 256 bytes. In this case, the time for transmitting 256 bytes is 204.8 ns. When the number of accelerator systems is 100, a transmission prompt message is transmitted about every 20 μs, and a processing request from the accelerator system can be transmitted at that interval. This means that once per second, the number of times that one accelerator system can perform uplink communication is 50,000 times.

  However, when the number of accelerator systems is 1000, the number of uplink communications per second is 5000. There is also a possibility that not all accelerator systems have information on upstream communication. An effective method in such a case is a multi-layered rhythm node indicated by “communication system”. An example is shown in FIG. FIG. 18 shows a basic configuration similar to that shown in FIG. 8, but the final-stage communication device has a different configuration. The final-stage communication devices (55, 56, 57) in FIG. 18 have functions of a dominant node and a tonic node, and further have rhythm nodes (21, 22, 23). The function of this final stage communication device is as follows. The final communication device 55 will be described as an example. The final-stage communication device 55 functions as a tonic node for the accelerator systems 1, 2, and 3. It also has a rhythm node. A transmission prompt message is transmitted from the rhythm node to each of the accelerator systems 1, 2, and 3. The accelerator system performs uplink communication after a predetermined waiting time after receiving the transmission reminder message. In this case, if there is no uplink information, for example, NACK is transmitted. In this way, the final stage communication device 55 collects and accumulates uplink communications from the accelerator systems 1, 2, and 3.

  In response to such an operation, the rhythm node 20 transmits a transmission prompt message using the final-stage communication devices 55, 56, and 57 as dominant nodes. This transmission reminder message is sent to a group intended for several stages of communication devices 55, 56, and 57. Further, it is preferable that the final communication devices 55, 56, and 57 do not transmit the transmission prompt message to the accelerator system. Even if it is sent, it is not addressed to the accelerator system and will be ignored, but it will reduce unnecessary traffic. As a dominant node, the final communication device performs uplink communication after receiving a transmission prompt message from the rhythm node 20 with a predetermined waiting time. Uplink communication may be performed one message at a time from each final stage communication device, but for example, 10 messages may be transmitted at a time. Of course, the predetermined waiting time becomes longer by the length of the entire message.

  By using such a method, it is possible to reduce empty uplink communication from the accelerator system. Furthermore, as shown in the “communication system”, when the number of uplink communications assigned to each dominant node is included in the transmission prompt message, a variable number of messages are uplinked from the final stage communication device. Furthermore, communication efficiency can be improved.

[Accelerator system failure]
By the way, when many accelerator systems are connected in this way, some accelerator systems may not be able to operate for some reason. Some reason may be due to a failure of the accelerator system, a failure of the branch line communication path, or a maintenance of the accelerator system. Thus, a countermeasure when a certain accelerator system does not operate will be described below.

  As a detection method when a certain accelerator system stops operating, a method is generally used in which a signal is periodically exchanged with the primary system, and if the signal is interrupted, it is regarded as a failure. In the present invention, the presence / absence of uplink communication for transmission prompting can be used for a signal, and it is reasonable to use it. In this way, when the primary system detects that the accelerator system is not operating, the primary system disconnects the accelerator system. On the primary system, back-out processing (returns the contents of the data to the contents at the start of the transaction) when the data in the database has been changed in a transaction that was processed by the accelerator system that became inoperative I do. If the backout content affects the accelerator system or a secondary system described later, the change information is transmitted to the accelerator system or the secondary system. In order to perform backout in this way, it is necessary to acquire the pre-change information (B log) of the relevant part when changing data in the primary system. The B log becomes unnecessary when the change in the primary system is completed and the change in the accelerator system or the secondary system is completed. However, there are cases where it can be used effectively by storing it for backup. This is explained in the secondary system section. From the change completion confirmation table, exclude the accelerator system that has become inoperable.

  Next, the primary system saves all subsequent change information including the change completion notice not received from the accelerator system. In addition, the application processing computer is prevented from sending a processing request to the inactive accelerator system, and the transaction at the time of inactivity is distributed to another accelerator system. This completes the separation of the accelerator system that has become inoperative. After that, the cause of the inoperability is searched manually, and the operation is restored so that it can be operated.

  When an accelerator system that has been out of service becomes operational, the following recovery measures are taken. From the primary system, the change information that has been saved is sent to the accelerator system that was out of service. Let this start time be t1. In the primary system, change information addressed to other accelerator systems is transmitted after time t1, but the accelerator system that has not been operated only receives and does not apply the changes. The accelerator system that was out of operation sequentially executes the changes based on the change information stored in the primary system, and sends a change completion notification to the primary system. The change completion notification may be performed every time the change of one piece of change information is completed, but may be performed collectively because the number of change completion notifications increases. Let t2 be the time when the saved change information disappears. After the time t2, the change is sequentially executed based on the change information saved by the accelerator system that has been inactive, and a change completion notice is transmitted to the primary system. The change completion notification may be performed every time the change of one piece of change information is completed, but may be performed collectively because the number of change completion notifications increases. Let t3 be the point in time when the change of the change information saved by the accelerator system that has been inactive has ended. At time t3, the accelerator system that was out of operation and the other accelerator system have been synchronized. Therefore, after t3, the accelerator system that has been inactive can be handled as an active accelerator system. In the primary system, the accelerator system that has become inoperable is excluded from the change completion confirmation table.

  Next, a description will be given of a case where the primary system is out of service. This will be described in detail in the description of the case where the secondary system is used.

[BUS type configuration]
Although the case of the cascade connection is described in FIG. 8, the configuration is possible even with the BUS type. FIG. 19 shows an example in which a BUS type configuration is used. The primary system and the accelerator system are connected by a BUS type communication path (41, 42). The communication path 41 is for downlink communication, and the communication path 42 is for uplink communication. Moreover, although both the communication paths 41 and 42 are one branch, the communication path 41 may be both branches. The reason why the communication path 42 is one-branch is that the upstream communication from the accelerator system does not flow downstream. When upstream communication from the accelerator system flows to the downstream side, for example, upstream communication from the accelerator system 3 also flows to the accelerator system 4 side, so that the accelerator systems 4, 5, There is a possibility that uplink communication from 6 and 7 may collide on the communication path 42. Such a collision can be avoided by leaving the communication path 42 in one branch and transmitting uplink communications from each accelerator system at appropriate intervals. An appropriate interval is nothing but taking a predetermined waiting time.

  A diagram relating to this predetermined waiting time is shown in FIG. Here, in order to simplify the drawing, four cases of accelerator systems 1 to 4 are shown. From the rhythm node (= primary system 0), a transmission reminder message is transmitted by broadcast. Then, unlike the case of the cascade-type transmission message, each accelerator system is sequentially increased in distance from the primary system, so it is delayed according to that distance and reaches each accelerator system. (S1). In the accelerator system, uplink communication is performed after a predetermined waiting time after receiving the transmission reminder message (S2, S3, S4, S5).

  The predetermined waiting time in this case is shown in detail in the “communication system”, but the equation when the distance between the accelerator systems is equal is as follows. The distance between the accelerator systems is D (meters), the message length is L (bytes), and the transmission speed of the communication path is S (bits / second). In contrast, the waiting time is T (nanoseconds). Also, the number of the accelerator system closest to the primary system is 1, and the nth dominant node is nth. Assuming that the waiting time of the nth dominant node is Tn (nanoseconds), Equation 2 representing the waiting time Tn is as follows.

  The case where the distance between each accelerator system is different is described in the “communication system”.

[Ring type]
FIG. 21 is a diagram in the case where an accelerator system is configured in a ring type. Here, in the communication path 41, for example, the branch line communication paths 331 and 312 are not disconnected and are connected. This enables broadcast communication and eliminates the problem that even if one of the nodes fails, the entire network stops. Here, the system consists of a primary system 0 and accelerator systems 1, 2, 3,... Since the communication path is a ring type, the communication path 41 is a single line. For this reason, in the communication channel 41, uplink communication and downlink communication are not separated. For convenience, communication from the primary system to the accelerator system direction is referred to as downlink communication, and communication from the accelerator system to the primary system direction is referred to as uplink communication, but communication flowing through the communication path 41 is uplink communication. Whether it is downlink communication depends on the destination. Each accelerator system is connected to the communication path 41 by branch line communication paths 311 and 312. Each branch line communication path is a single branch, the branch line communication paths 311 and 321 are used for receiving downlink communication, and the branch line communication paths 312 and 322 are used for transmitting uplink communication.

[Uplink communication]
Here, it is assumed that a rhythm node 20 is provided in the primary system 0. First, for uplink communication, a transmission reminder message is transmitted from the rhythm node 20 by broadcast. After receiving the transmission reminder message, the accelerator system transmits the upstream message with a predetermined waiting time according to the distance between the accelerator systems. For the predetermined waiting time in this case, an expression in the BUS type communication network is used. Further, in the case of the ring type, the message addressed to the primary system may be used without being limited to uplink communication. In the primary system, a message addressed to itself is received, and other messages are transmitted again as downlink communication to the communication path, but do not exist in the accelerator system. The primary system discards all received messages other than those addressed to itself.

[Downlink communication]
Next, the rhythm node assigns downlink communication to the primary system 0. Therefore, the primary system 0 performs downlink communication. All communication messages returned to the primary system through downstream communication are discarded.

Claims (5)

A data record having a data item including a primary key, a primary block for storing the data records in the order of the primary key, and a location table having a location table entry in which the address of each primary block is described in a continuous area A primary system with
An accelerator system having a Fland location table having a Fland location table entry in which the address of each primary block is written in a continuous area;
A communication path,
When the change information is sent from the primary system by broadcast and the change information is received by the accelerator system, the change completion notification for the primary system is sent from the accelerator system after a predetermined waiting time. A computer system characterized by A data record having a data item including a primary key, a primary block for storing the data records in the order of the primary key, and a location table having a location table entry in which the address of each primary block is described in a continuous area A primary system with
An accelerator system having a Fland location table having a Fland location table entry in which the address of each primary block is written in a continuous area;
A communication path,
When the rhythm node sends a transmission dunning message in a broadcast and the accelerator system receives the transmission dunning message, the transmission from the accelerator system to the primary system is performed after a predetermined waiting time. Characteristic computer system. Consecutive sequence of data records with data items including primary key and alternate key, alternate key entry consisting of alternate key and primary key, alternate key block including alternate key entry, and alternate key location table entry A primary system holding an alternate key location table in the region;
An accelerator system having a franc alternate key location table having a franc alternate key location table entry in a continuous region;
A communication path,
The primary system sends alternate key location table change information, and each accelerator system receives the alternate key location table change information, modifies the alternate key land location table, and replaces the alternate key. A computer system characterized in that a change completion notification is transmitted from the accelerator system to the primary system after a predetermined waiting time when receiving the change information of the location table. Consecutive sequence of data records with data items including primary key and alternate key, alternate key entry consisting of alternate key and primary key, alternate key block including alternate key entry, and alternate key location table entry A primary system holding an alternate key location table in the region;
An accelerator system having a franc alternate key location table having a franc alternate key location table entry in a continuous region;
A communication path,
When the rhythm node sends a transmission dunning message in a broadcast and the accelerator system receives the transmission dunning message, the transmission from the accelerator system to the primary system is performed after a predetermined waiting time. Characteristic computer system. A data record having a data item including a primary key, a primary block for storing the data records in the order of the primary key, and a location table having a location table entry in which the address of each primary block is described in a continuous area A primary system with
A plurality of accelerator systems having a land location table having a land location table entry in which the address of each primary block is written in a continuous area;
A communication path;
A rhythm node that transmits a transmission reminder message,
The rhythm node sends a transmission dunning message to each accelerator system, and each accelerator system sends a processing request to the primary system after a predetermined waiting time triggered by receiving the transmission dunning message. A computer system characterized by

Images