JP2811916B2 - Data file access method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Summary] The present invention relates to a data file access method for efficiently searching and updating data in a database, and to provide an optimum access method to a divided memory device such as a file disk. In a device for storing a predetermined data file in a plurality of memory devices and retrieving the data file, the number n of attributes to be processed and the number of divisions of an attribute domain which is a prime number or its exponent q, and a code parameter generating means for inputting the number L of memory devices to obtain a predetermined code parameter; and a parity check matrix generating means for generating a parity check matrix H based on an output of the code parameter generating means. The number of the memory device in which the set of records referred to as a bucket is stored depends on the code parameter generation method. Give storage number of the memory device by the attribute vector u obtained from the a matrix H records generated based on the output, the code parameter generating means, the number of memory devices L (L
= Q ^r ), the number of the attributes is code length n,
A code C on a finite field GF (q) having the number of information symbols k (= n−r) is selected and the code C is generated. A maximum distance separation code, a p-double error correcting BCH code or a shortened code thereof ,
In this configuration, any one of a code obtained by connecting several Hamming codes and a shortened Hamming code, a shortened Hamming code, an extended Hamming code, an even-weight Hamming code, and an expanded Hamming code is used.

[Industrial applications]

The present invention relates to a data file access method for efficiently searching and updating data in a database.

A question for searching from a record file having a specified value for a specified attribute is called a partial matching question. Hereinafter, the partial matching question may be simply referred to as a question. Generally, for a file consisting of records having more than n attributes X1, X2,..., Xn, the question is written in the form of Q: (X1 = a1,..., Xn = an). Here, each ai is a value belonging to the domain of the attribute Xi. If * is specified as ai,
Indicates that the attribute value is not specified. As a special case, when '*' is written for all attributes, all records belonging to the data file are searched.
When there is no '*' (ie, when all the attribute values are specified), it is checked whether the specified record exists. For example, in a data file consisting of four attributes, subject, day of the week, teacher, and subject, if you want to know the subject that the information engineering teacher is in charge of on Wednesday and the teacher, Q: (subject = *, day = ' (Water ', Teacher =' * ', Department =' Information Engineering ')

The partial matching query is the most basic operation for searching and updating data in the database, and it is important to improve the processing efficiency. If the attribute for which the value is specified in the question is determined to some extent and the number is small,
By creating an index on those attributes, questions can be processed efficiently. However, when the specified attributes are often unknown, it takes time to maintain the index.

Conventionally, in order to solve such a problem, many methods for efficiently processing a question using a direct product file have been studied. The direct product file is a file that is divided in units of buckets that can be taken out to the main memory of the computer by a single reading from an external storage such as a disk that can be accessed randomly. Now, the domain Di of each attribute Xi is _Di1 , _Di2 , ...
… _Assuming that it is divided into D _iq , each bucket is D _1j1 × D
_The bucket, which is a subset of _2J2 ×... × D _njn , has length n
Is represented by a q-element vector (j ₁ , j ₂ ,..., J _n ). Hereinafter, this vector is called an attribute vector.

In the direct product file, a bucket including a record that satisfies the condition specified by the question is obtained in a fixed time. Hereinafter, the direct product file is simply referred to as a file.

In an environment where a plurality of disks (memory devices) that can be accessed in parallel, such as a large-scale database system, are used, query processing can be made more efficient by storing buckets belonging to a file in a distributed manner on those disks. Here, it is assumed that the other processing time in the computer is sufficiently shorter than the disk access time. In this case, the processing time of the question can be evaluated by the maximum value of the number of buckets read from each disk.

Therefore, when buckets to be read for a query are stored equally on all disks, the query can be processed most efficiently. At this time, the disk layout of the file is optimal for the question.

FIG. 4 shows a system block diagram on which the present invention is based. When an attribute vector of a file to be extracted in response to a query is output from the computer 301, disk numbers 1 to L are calculated by the file storage device 302, and the files are read from the disks 303 to 305 in parallel through data I / O. You. The same processing is performed for storing a disk.

The present invention relates to an access method for a divided memory device such as a disk of the above-mentioned file, and provides a configuration for outputting a disk number for optimally arranging a file in response to an arbitrary question.

[Conventional technology]

In a conventional method of arranging files on a disk, a method of arranging files on a disk using an error correction code has been proposed. FIG. 5 shows a block diagram of a conventional system.

In the conventional method, first, the number of attributes n and the number of disks L
Is input, the logarithmic calculation circuit 401 calculates [log ₂ L = r], and the subtractor 402 calculates the output r of the logarithmic calculation circuit 401 from n.
And the result is set to k. Number of attributes n
Is input to the (n, k) code address generator 403.
The (n, k) code address generator 403 generates a storage address of a code database having a code length n and the number of information symbols k. The code parameter database 404 outputs the minimum distance of the (n, k) code to the maximum value determiner 405. When the addresses of all the (n, k) codes are completed, the maximum value judging unit 405 generates an address of the code with the minimum distance having the maximum value, and (n, k)
Input to the code address generator 403. The (n, k) code address generator 403 outputs the output address of the maximum value determiner 405 to the code parameter database 404. At this time, the code parameter database 404 stores the check matrix generation circuit 406
Then, data necessary for generating a parity check matrix is output.

The check matrix H generated by the check matrix generation circuit 406 is input to the matrix operation circuit 407, and an attribute vector (u) having a length n (the number of attributes) and a matrix operation H · u of the matrix H are performed. Output as the storage device number.

[Problems to be solved by the invention]

In the conventional method, it is necessary to search for a parity check matrix of which code to use every time a parameter is given. In addition, it has been proved that the optimal disk layout of the file is only when the number of '*' appearing in the question is equal to or less than the minimum Hamming distance of the code. (References C. Faloutsus and D. Metaxas, “Declurtering Using
Error Correction Codes, "Proc.of the 8th ACM Symp.
on Principles of Database Systems, "pp. 253-358, Mar
ch 1989). However, under this condition, there is virtually no '*' in the question, that is, it includes only the case where values must be specified for most attributes, which limits its practical application. There is.

An object of the present invention is to provide an optimal access method to a divided memory device such as a file disk.

[Means for solving the problem]

1 (a) and 1 (b) are diagrams for explaining the principle of the present invention.
(A) is a principle configuration diagram of the present invention. In the figure, the number of attributes
The inspection circuit 101, which receives n, the number of storage devices L, and the number of divisions q of the domain of the attribute, as inputs, searches for which code selection condition is satisfied by n, L, q, and calculates parameters. Where NA
Is the range where '*' can be optimized.

The present invention relates to an apparatus for storing a predetermined data file in a plurality of memory devices and searching for the data file.
Code parameter generation means 101, 102, 103 for receiving predetermined numbers of code parameters by inputting the number n of attributes to be processed, the number q of divisions of the attribute domain which is a prime number or its exponent, and the number L of memory devices, A parity check matrix generation unit 104 that generates a parity check matrix H based on an output of the code parameter generation unit, and a number of the memory device in which a certain record (or a set of records called a bucket) is stored, A matrix H generated based on the output and an attribute vector u obtained from the record
(A q-element vector of length n) to obtain the storage number of the memory device, and the code parameter generation means, when the number of memory devices is L (L = ^qr ),
The number of the attributes is defined as a code length n, and the number of information symbols k (= n−
r), a code C on the finite field GF (q) is selected.
Is generated. That is, the code address generator 10
2 outputs the address of the code stored in the code parameter database 103 based on the parameters from the check circuit 101 for n, L, and q. The parity check matrix generation circuit 104 generates a parity check matrix based on the code data from the code parameter database 103. The parity check matrix H generated by the parity check matrix generation circuit 104 is input to the matrix calculation circuit 105, and a matrix calculation H × u corresponding to code syndrome calculation is performed with respect to the input of the vector (u) of the value of the attribute of length n. The result is output as a storage number.

The flow of the inspection process of the inspection circuit 101 shown in FIG.
It is shown in FIG. In FIG. 1 (b), the number n of attributes and the number L of disk devices as memory devices (=
q ^r ). Here, q is a prime number or a power of a prime number.

First, in q is a power of 2, n = q + 2, L = q 3 or, q is a power of a prime number, if the n ≦ q + 1 and 1 ≦ r <n,
If the maximum distance separating code 11 that satisfies this condition is selected as the code C, the q-element code 12 is selected as the code C if q is neither a power of 2 nor a power of a prime number and is a number other than 2.

Also, when q is 2, if the range NA that can be optimized for '*' is less than half the code length n of the attribute, if p is an integer of 2 or more, r is a multiple of p And n ≦ 2 ^{r / p} −
When 1, the p-error correction BCH code or its shortened code 13 is used as the code word C.

In addition, there is no appropriate BCH code parameter, that is, r
Is a multiple of p, and there is no p that satisfies n ≦ 2 ^{r / p} −1 or the range NA where “*” is to be optimized is larger than half the number of attributes. When the number of attributes is larger than n, a code 14 in which some Hamming codes and shortened Hamming codes are connected is selected as the code C. When n = L-1, the Hamming code 15 is used and L / 2 <
If n <L-1, the shortened Hamming code 16 is used, and n = L / 2
In the case of, the extended Hamming code 17 is used, and when n = L / 2-1, the even weighted Hamming code is used, and r + 1 ≦ n <L / 2−1.
In the case of, the shortened code of the extended Hamming code is used as the code C.

[Action]

In the present invention, by checking the parameters of codes that can be taken by the n, L, q check circuit 101, it becomes possible to eliminate the need to search for all codes having the parameters. This is because, while the condition for selecting a code other than the minimum distance was not clear conventionally, the present invention can clarify the condition of the code which is optimal even when the number of '*' appearing in the query is large, and the parameter is Once determined, the appropriate code will be determined.

〔Example〕

First Embodiment FIG. 2 is an embodiment of the present invention when q = 2.

The input is the number of disk devices L (= 2 ^r ) selected to be a power of n, the number of attributes, and 2. When n is entered, n,
The L, q inspection circuit 201 sequentially compares the magnitude relations of n, L, and NA, and determines the code according to the flow of FIG. NA
A magnitude comparison circuit 203 compares the magnitudes of the calculation circuits 202, n and L, and their square roots, etc., with the magnitude r, a sign determination circuit 205 determines the sign from the comparison result, and a control circuit 204 controls the whole.

When the code is determined, an address of the memory 207 in which the code parameter is stored is generated by the address generation circuit 206. The memory 207 stores parameters necessary for generating a parity check matrix of a code determined by n and L.

The parameters stored in the memory 207 are input to the parity check matrix generation circuit 208. The input data is stored in the memory 209 under the control of the control circuit 212, and is stored in the selector 210.
Sets the tap coefficient of the shift register 211.
The check matrix is sequentially calculated by the shift register 211 and is stored in the memory 2
Store to 09. When the calculation is completed, the selector 210 is switched and the output of the memory 209 is switched to the matrix operation circuit 213 to output the check matrix H. The parity check matrix calculation circuit 213 writes the parity check matrix into the register 214. The matrix operation circuit 213 receives the attribute information (vector u of length n) and performs a matrix operation with the parity check matrix H and an exclusive OR calculation for each element in the EOR circuits 215 to 217.
The exclusive OR circuit 218 performs an exclusive OR operation on the result again, and outputs the disk number of the stored or read file.

The following is a numerical example of n = 63, L = 64. When this data is input, a Hamming code (15 in FIG. 1 (b)) is selected because n = L-1. At this time, from the memory 207,
The generator polynomial X ⁶ + X ⁵ + X ² + X + 1 of this code is output,
The parity check matrix H shown in FIG. 3A is generated by the parity check matrix generation circuit 208.

At this time, the number of attributes n is 63, and when the attribute u of a certain record is the transposed matrix u ^T in FIG. 3B (where the subscript T indicates a transposed matrix), The disk number to be stored is (Hu) ^T = [110111] = 55, and is stored in the 55th disk.

Description of Other Embodiments Although the above embodiment has been described with reference to a binary code, it can be easily extended to a p-code. However, in this case, all codeword operations are modq operations.

In addition, for the code selection, the n, 1, q check circuit checked all codes. However, the range of the number of optimal '*' s was limited to the case where the number of attributes was small. BC
By using only the H code, the code database can be reduced and the configuration of the check matrix generation circuit can be simplified. In addition, the range of the number of optimal “*” is limited to the case where the number of attributes is large, and only the Hamming code is selected, thereby reducing the code database and simplifying the configuration of the parity check matrix generation circuit. It is also possible.

〔The invention's effect〕

As described above, according to the present invention, it is possible to eliminate the code check by checking the code parameters that can be taken by the n, L, q check circuit.

Furthermore, especially when the maximum distance separating code C is selected,
The disk layout of the file is optimal for the number and all of the '*' that appear in the question.

Also, in the case of the Hamming code in the case of q = 2, the minimum distance is 3 (4 in the case of expansion), so that the disk arrangement of the file is optimal when the number of “*” appearing in the question is 3 or 4 or less. In addition, the number of '*' s that is about half or more of the code length is optimal, and the optimal range of the number of '*' s is wider than that of the conventional selection method. A code C having a selection range of 'can be obtained.

[Brief description of the drawings]

FIG. 1 is a diagram for explaining the principle of the present invention. FIG. 1 (a) is a diagram showing the principle configuration, FIG. 1 (b) is a flowchart of an n, L, q test circuit processing, and FIG. FIG. 3 is a diagram showing an example of a parity check matrix and an attribute.
(A) is an example of a parity check matrix output from a parity check matrix generation circuit, (b) is an example of an attribute, FIG. 4 is a system configuration diagram as a background of the invention, and FIG. 5 is a conventional configuration diagram It is.

──────────────────────────────────────────────────続 き Continued on front page (58) Field surveyed (Int.Cl. ⁶ , DB name) G06F 12/00 G06F 17/30

Claims (4)

(57) [Claims] An apparatus for storing a predetermined data file in a plurality of memory devices and retrieving the data file, wherein a number (n) of attributes to be processed and a domain of an attribute which is a prime number or an exponent thereof. The code parameter generation means (101, 102, 103) which receives the number of divisions (q) and the number of memory devices (L) as input and obtains predetermined code parameters, and the check matrix (H) based on the output of the code parameter generation means
And a check matrix generation means (104) for generating the number of the memory device in which a certain record (or a set of records referred to as buckets) is stored, wherein a number generated based on the output of the code parameter generation means (H) and an attribute vector (u) obtained from the record, wherein a storage number of the memory device is obtained. 2. The method according to claim 1, wherein when the number of memory devices is L (L = q ^r ), the code parameter generation means sets the number of the attributes to a code length n and the number of information symbols k (= n−r). A finite field
A code C on GF (q) is selected to generate the code C, where q is a prime number or a power thereof, n ≦ q + 1 and 1 ≦ r <
n or, q is a power of two, n = q + 2, L = claims a maximum distance separable code in the case of q ^3, characterized in that selected as the code C (1) Data file access method according. 3. When the number of memory devices is L (L = q ^r ), the code parameter generation means sets the number of the attributes to a code length n and the number of information symbols k (= n−r). A finite field
A code C on GF (q) is selected to generate the code C. When q = 2, when p is an integer of 2 or more, r is a multiple of p and n ≦ 2 ^{r / p} −1 2. The data file access method according to claim 1, wherein a p-type error correcting BCH code or its shortened code is used as the code word C in the case of (1). 4. When the number of memory devices is L (L = q ^r ), the code parameter generating means sets the number of the attributes to a code length n and the number of information symbols k (= n−r). A finite field
A code C on GF (q) is selected to generate the code C. When q is 2, when the number of attributes n is larger than the number of disks L, some Hamming codes and shortened Hamming codes are used. Is selected, a Hamming code is used when n = L-1, a shortened Hamming code is used when L / 2 <n <L-1, and an expansion is used when n = L / 2. It is characterized in that a Hamming code is used, an even-weighted Hamming code is used when n = L / 2-1, and a shortened code of the extended Hamming code is used as a code C when r + 1 ≦ n <L / 2-1. The data file access method according to claim 1, wherein