JPH103458A - Load balancing method for database management system - Google Patents

Abstract

[PROBLEMS] To search databases in a system in which databases 14, 24, and 34 are connected to processing devices 1, 2, and 3, respectively, and perform database management by database management systems 11, 21, and 31, respectively. The following table join processing is load-balanced based on the size of the unprocessed load. SOLUTION: The processing line number counters 12, 22, 32 include:
The accumulated number of unprocessed lines updated according to the number of processed lines of the combining process is held. After the search processing of the tables x and y is completed, the database management system 11 selects a processing device so as to reduce the data transfer load of the network 10 due to the combining process when allocating the processing device to perform the combining process. . When the network loads of the two processing devices are equal, a processing device with a small load is selected based on the values of the processing line number counters 12, 22, and 32.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a load distribution method for a database management system, and more particularly to a method for distributing the load of an information processing device in a system in which a plurality of information processing devices are interconnected via a network.

[0002]

2. Description of the Related Art In a system in which a plurality of processing devices are connected to each other via a network, a CP of each processing device is used.
A method of measuring U utilization and distributing the load according to the CPU utilization has been performed. In other words, this method uses the C at the time of assigning a new job or process to any processing device.
The load on the processing device is distributed by selecting the processing device with the lowest PU utilization. However, in this method, the magnitude of the load applied to the processing devices is estimated based on the CPU utilization at a point in time, and the method necessarily reflects the magnitude of the load actually held by each processing device. I can't say that.

[0003] Related to this kind of technology are, for example, JP-A-4-129399 and JP-A-5-1993.
No. 151177.

[0004]

Since the CPU utilization does not necessarily reflect the magnitude of the load that the processing unit must digest in the future, if the load distribution control is performed according to the CPU utilization, the load cannot be correctly distributed. In some cases, processing is concentrated on one processing device.

An object of the present invention is to solve the above problem by performing load distribution control in a database management system in accordance with the size of the unprocessed load.

[0006]

According to the present invention, there is provided a system in which a database composed of at least one table is connected to each processing device and the database is searched in parallel. The cumulative number of processed lines indicating the number of unprocessed processes is constantly updated according to the number of processed lines at the time of performing the combining process, and when the network load generated by the combining process for the two processing devices is equal. The present invention is characterized by a load distribution method of allocating a combining process to a processing device having a small number of accumulated processing rows. The reason for adopting the number of processing lines of the combining process as an index indicating the load of the processing device is that the CPU time required for this processing can be predicted from the number of processing lines of the combining process, and the magnitude of the load held by the processing device can be almost grasped. That's why.

[0007]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a configuration diagram of a parallel computer system according to the present embodiment. The parallel computer system includes a plurality of processing devices 1, 2, 3, and a network 10 interconnecting these processing devices. Each processing device 1, 2, 3
Connect the databases 14, 24, 34 respectively. These databases are relational databases. The database 14 is composed of tables A and B, the database 24 is composed of tables C and D, and the database 34 is composed of table E. Database management system 11, 21, 3
Numeral 1 is a program for managing the databases 14, 24, and 34, respectively, and has processing line number counters 12, 22, and 32 for storing the cumulative number of processing lines for unprocessed data when combining search results. UAPs (user application programs) 13, 23, and 33 are programs that issue database search requests to the database management systems 11, 21, and 31, respectively. The database management table 15 is a table that stores the correspondence between the names of the tables that make up the database and the processing devices that own the tables, and is stored on a storage device connected to the processing device 1. Although not shown, the database management table 15 is also connected to the processing devices 2 and 3, respectively.

[0009] When the UAP 13 retrieves the table x and the table y and issues an SQL that joins the tables, the database management system 11 refers to the database management table 15 to execute the processing of the processing device that owns the tables x and y. Find the number,
A search request is issued to the database management system of the processing device having each table. Each database management system searches each table x and table y, reports the search row number x _n and y _n hit when the search process has been completed in the database management system 11. The database management system 11
A table join processing request is issued to one of the processing apparatuses having the table x or the table y so as to reduce the data transfer amount of the search result and the join result via the network 10. The database management system of the processing device that has received the join processing request requests the search result from the database management system of the processing device that owns the other table, and when receiving the search result, combines the search result with the processing row number counter 12, 22, or 32. The combining process is performed after the number of processing lines is added. When the combining processing is completed, the number of combined processing rows is subtracted from the number of processed rows counter, and then the combining result is transmitted to the database management system 11, and the database management system 11 passes the combining result to the UAP 13. When the transfer between the processing devices becomes equivalent irrespective of which of the processing devices owning the table x or the table y is selected, the database management system 11 sets the processing line of the processing line number counter of each processing device having the table x and the table y. The numbers are compared, and a table join request is issued to the database management system of the processing device with a small load.

FIG. 2 is a diagram showing a data format of the database management table 15. The database management table 15 holds correspondence between table identifiers and numbers of processing apparatuses having the tables.

FIG. 3 is a flow chart showing the flow of the processing of the database management system 11 after the SQL for performing the joining of the tables after the tables x and y are searched from the UAP 13 is issued. However, steps 41, 43, and 45 may involve data transfer of the processing device via the network 10 and processing of another database management system. The database management system 11 refers to the database management table 15 to obtain the numbers of the processing devices that own the tables x and y, and executes a search process (step 41). That is, a search request is transmitted to the database management system of the processing device that owns the table x and the table y, specifying the transaction ID and the table x or the table y, respectively. If the processing device 1 owns one or both of the table x and the table y, it only issues a search request to the search processing program in the database management system 11. When the search process is completed, the database management system 11 are the same transaction ID, the number of rows x _n which is hit as a result of the search, receives the value of y _n and processing line number counter. Next, if the table x and the table y are the same processing device (YES in step 42), the database management system 11 performs a joining process by a processing device having the table x (step 43). When the processing devices that own the table x and the table y are different (step 42 NO), the number of search rows that hit the table x and the table y is compared (step 44).
If the number of rows to be searched for in the table x is large (x _n > y _n ), a combining process is performed by a certain processing device in the table x (step 43). Table y
If the number of search rows is large (x _n <y _n ), the combining process is performed by a processor having a table y (step 45). When the number of search rows in table x is equal to the number of search rows in table y (x _n = y _n ), U
If the AP 13 and the table x are in the same processing device (step 46)
YES), a combining process is performed by a processing device having a table x (step 43). If the UAP 13 is different from the processing device having the table x (NO in step 46), and if the UAP 13 and the table y are in the same processing device (YES in step 47), the combining device is processed by the processing device having the table y (step 45). UA
If P13, table x, and table y are in different processing devices (step 47 NO), the database management system 11 processes the number of rows p (x) _{m of the} processing device having table x and the processing of the processing device having table y. The number of rows p (y) _m is compared (step 48). If the number of processed rows p (x) _m is smaller or equal (p (x) _m ≤ p (y) _m ), a combining process is performed by a processing device having a table x (step 43). Number of processed rows p (y)
_{If m} is smaller (p (x) _m > p (y) _m ), a combining process is performed by a processing device having a table y (step 45).

FIG. 4 is a flow chart showing the processing of combining steps 43 and 45 further developed.
The database management system 11 of the processing apparatus 1 having the UAP 13 transmits a join processing request to the processing apparatus having one table and performing the join processing by specifying the transaction ID at the time of search and the tables x and y (step 51). ). When the processing device that performs the combining process receives the request for the combining process (step 52), it specifies the transaction ID at the time of the search and transmits a search result request to the processing device having the other table (step 53). When the processing device having the other table receives this search result request (step 54), it specifies the transaction ID and transmits the search result and the number of search rows that have been hit to the processing device that performs the combining process (step 5).
5). Upon receiving the search result (step 56), the processing device that performs the combining process updates the processing row number counter by adding the number of processing rows of the combining process to the processing row number counter (step 57). A joining process is performed on the search result of the table y (step 58). When the combining processing is completed, the processing row number counter is updated by subtracting the number of processing rows of the combining processing from the number of processing row counters (step 59). Next, the processing device that performs the combining process is the processing device 1
A transaction result is transmitted by designating the transaction ID at the time of the request for the coupling process (step 60). The database management system 11 receives this connection result (step 6).
1) Pass it to the UAP 13 (step 62). UAP13
If a certain processing device 1 and a processing device that performs a connection process are the same processing device, the processing of the same processing device becomes a process in the database management system 11, and there is no transfer of a connection processing request and a connection result via the network 10. . If the processing device performing the combining process and the processing device having the other table are the same processing device, the processing of the same processing device is a process in the same database management system, and the search result request and the search result No transfer. FIG. 4
In the example, the number of search rows that hit the number of rows processed by the join process x _n
And y _n , but other than x _n and y _n such as the product of x _n and y _n
Anything that can be expressed by the function of _n may be used.

According to the above embodiment, since only the database management system that manages the table to be searched can execute the search processing, the database management system 11 issues a search request to the processing device that manages each table. If table x and table y are in the same processing device (step 42 YES),
Since the load on the network 10 for transmitting the search result is not applied, the combining processing may be performed by the same processing device having the tables x and y. When the processing device that owns the table x and the table y are different (step 42 NO), the number x _{n of} search rows that have been hit is obtained.
And since the binding processing by the processing device with the larger number of search lines of y _n, load on the network 10 is small because of the search result transfer. When the number of search rows x _n and y _n are the same (x
_{In n} = y _n ), the connection processing is performed by a processing device in which the UAP and one of the tables are in the same processing device, so that the load on the network 10 is not applied to transfer the connection result. Generally, it is considered that the data amount of the combination result is smaller than the data amount of the search result. The number of searched rows x _n and y _n are equal and U
If the AP, the table x, and the table y are in different processing devices (NO in step 47), the network may be processed by the processing device having the table x or by the processing device having the table y. Since the load of the processing device 10 is the same, the database management system 11 compares the contents of the processing line number counter of the processing device having the table x with the contents of the processing line number counter of the processing device having the table y. A connection processing request is issued to the processing device with the smaller load. It can be considered that the load on the processing device that performs the combining process increases in accordance with the number of rows processed in the combining process.

In the above embodiment, the UAP is located in any one of the processing devices.
3 may be in a client device connected to the network 10. At this time, the processing devices 1, 2, and 3
Functions as a server that manages a database. Processing devices 1 via a network 10 from a client
When the SQL is issued to 2 or 3, the database management system of the processing device that has received the SQL performs the above-described processing shown in FIGS. Send the join result to the client.

[0015]

According to the present invention, it is possible to predict the magnitude of the unprocessed load held by each processing device by constantly updating the cumulative number of processing lines of the combining process. By allocating the combination processing to the processing devices, it is possible to prevent the concentration of the load, which has been a problem in the related art.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a parallel computer system according to an embodiment.

FIG. 2 is a diagram illustrating a data format of a database management table 15 according to the embodiment.

FIG. 3 is a flowchart illustrating a flow of a process of distributing a load of a combining process according to the embodiment.

FIG. 4 is a flowchart showing an expansion of the combining process of FIG. 3;

[Explanation of symbols]

1, 2, 3: processing device, 11, 21, 31: database management system, 12, 22, 32: processing row counter, 14, 24, 34: database

Claims (1)

[Claims] A plurality of information processing apparatuses are mutually connected via a network, a database composed of at least one table is connected to each information processing apparatus, and a database management system in each information processing apparatus is connected. A load balancing method for a system in which any database management system performs a process of joining tables of search results so as to perform a process of searching a table in a database to be searched and reduce a data transfer amount on the network. The apparatus always updates the cumulative number of unprocessed lines in accordance with the number of processed lines of the combining process, and when the data transfer amounts of the two information processing apparatuses are equal, the information processing apparatus with the smaller number of cumulative processed lines A load management method for a database management system, wherein