This invention relates to an indexing method for operating a database management system (DBMS) whose data area was partition.
The DBMS usually uses an index in order to speed up retrieval. Especially the B-Tree structure and its extended structure are data structures suitable for a disk unit, and therefore are widely used as the most common index structure. In the B-Tree structured index is obtained by sorting a specific one or plurality of columns in a table in advance. The index is arranged so that search efficiency is increased, whereby search with the column being used as a key is speeded up.
Meanwhile, as the amount of data has increased in recent years, attentions have been attracted to the cluster DBMS that uses a plurality of DB servers. As one method for operating the cluster DBMS, there is a method in which partition is performed with the hash-division algorithm or the range division method using a specific one or plurality of partition key columns in a table so that the same row in the table is not shared by the DB servers.
The cluster DBMS being operated by the above-mentioned method comes with a disadvantage that it will take a time to alter its configuration, such as a change of the number of DB servers. With the cluster DBMS, in the case of query processing that requires search results of all the DB servers to be summarized, query processing performance is governed by a DB server that takes the largest time. Since occurrence of a difference in the amount of data among the servers generally degrades the performance, it is essential to equalize the amount of data when altering the configuration. However, this processing requires data on a disk unit to be transferred to a physically different place, therefore taking a long time to do so.
U.S. Pat. No. 5,551,027 gives description about a hierarchical index for narrowing retrieval by a non-partition key in a cluster environment of the non-shared type DBMS. According to it, by using a global index consisting of a pair of a non-partition key and server number that maintains its column, it is possible to narrow down a server to be used for retrieval. However, the global index has only that capacity, but not a capacity for indicating a specific column in the table.
The inventors of this invention have previously proposed an invention in which a sharable storage is partitioned in advance and only allocation thereof is altered at the time of configuration alteration of the DBMS, whereby high-speed configuration alteration that does not accompany transfer of data is made feasible (JP 2004-003601 A).
In addition, it is also possible to imitate a function of the method of the prior application only using partitioned storages, without using the allocation management table. As one example, the imitating is possible by creating a large number of small LU's each corresponding to a small area in the shared storage on a network and manually managing allocation of these LU's. However, in this case, it is necessary to secure all matters that the DBMS manages, such as consistency management of transaction, which makes the operation complicated.
However, in the case where JP 2004-003601 A and the imitated technique described above are applied to a DBMS equipped with an index search function, there arises a problem that the index needs to be re-created. This is a problem arising from the DBMS having a configuration that makes it possible to create an index by specifying a column different from a partition key column of the table as a key. Re-creation of the index generally requires the whole table to be scanned, which needs a considerable time. Because of this, in the case of presence of the index, a merit of the high-speed configuration alteration by JP 2004-003601 A and the imitated technique described above will become faded.
There is also a case, as a solution method enabled by the conventional technique, where an index area 301 is previously partitioned (302) by the same dividing method as that of the table area 204 into the same number of partitioned areas, as shown in
Therefore, the problem that is intended to be solved by this invention is concerned with an operation of a database management system that consists of a plurality of database servers and enables the database servers to transfer data between them by partitioning a data area into small areas and altering allocation of the small areas to the database servers. More specifically, the problem is the occurrence of degradation in processing speed that accompanies the re-creation of the index after the configuration alteration. If this problem is intended to be solved by using the conventional technique, the problem is an occurrence of noticeable degradation in processing speed to a specific query at the time of a steady state operation.
In the operation of a database management system that consists of a plurality of database servers and enables the database servers to transfer data between them by partitioning a data area into small areas and altering allocation of the small areas to the database servers, indices created for the respective small areas and an index for all the small areas allocated to each database server are used in combination.
In the operation of a database management system that enables the database servers to transfer data between them by partitioning a data area into small areas and altering allocation of the small areas to the database servers, after the configuration alteration by altering allocation of the partitioned data areas, noticeable degradation in processing speed caused by re-creation of the index will not occur and, at the time of a steady state operation, noticeable degradation in processing speed to a specific query will not occur.
In this embodiment, the function shown in JP 2004-003601 A is applied to the table area 508 and the index area #1509. The scheduler module 505 is made to perform schedule management of the configuration alteration, and the mapping module 506 is made to perform allocation management of the areas 508,509 on the shared storage. In order to take such a configuration, it is necessary to guarantee that the table area 508 and the index area #1509 will establish one-to-one correspondence. In this embodiment, the hash-division algorithm was used to partition the table, and the B-Tree index was chosen to be the index. Further, using the technique shown in
Note here that when the configuration is altered, the small areas allocated to the back end modules 501,502 as a table area will change; therefore, discrepancy of contents will arise between the table area 508 and the index areas #2510, 511. In order to solve the problem, it is necessary to manage the large index before and after the configuration alteration.
Note that, in order to re-create the large index by the conventional technique, all the items in a pertinent table need to be scanned. However, in the environment to which this invention is applied, it is guaranteed that the small index corresponding to the large index surely exists and the small index has already been sorted. Therefore, speed-up in re-creating the large index can be attained by creating the large index from the small index, not from the table.
This embodiment imitates the function shown in JP 2004-003601 A without using the allocation management table. In this embodiment, the table area and the index area #1 in the shared storage are divided into 12 virtual volumes 705, 706, respectively, and allocation of these virtual volumes 705, 706 to the back end servers 701, 702 is managed manually, whereby the functions shown in JP 2004-003601 A are imitated. Storing of the table in the 12-divided virtual volumes 705 is done by taking a method whereby a table divided into to create 12 parts similarly by the hash-division algorithm. Here, a divided index that is created in the index area #1706 is also referred to as a small index. In the other index areas #2707,708, an index that corresponds to all the virtual volumes that are allocated to the back end modules 701, 702 is created. The index created in the index areas #2707, 708 is referred to as a large index.
In this embodiment, just like the first embodiment, when the configuration is altered, there will occur discrepancy in contents between the table area and the index area #2.
Number | Date | Country | Kind |
---|---|---|---|
2005-202401 | Jul 2005 | JP | national |
This application is a continuation application of U.S. Ser. No. 11/483,548, filed Jul. 11, 2006, the entirety of which is incorporated herein by reference. The present application claims priority from Japanese application JP 2005-202401 filed in Jul. 12, 2005, the content of which is hereby incorporated by reference into this application.
Number | Date | Country | |
---|---|---|---|
Parent | 11483548 | Jul 2006 | US |
Child | 12574398 | US |