JPS62189539A - System for evaluating performance of related data base managing system - Google Patents

Abstract

PURPOSE:To evaluate the performance of a vector type related data base manag ing system and to previously predict the effect of hardware by giving a means outputting information executing a vector instruction to a file, and a means which uses the contents of the file and evaluates the performance. CONSTITUTION:With a trace file 5 as an input the related data base managing system 11 executes a data load function, and generates a trace table T1 in th data base 13, where a hardware table T2 is generated. With the aid of an analysis command inputted through a terminal 14 the related data base managing system 12 analyzes performance using the trace table T1 and the hardware table T2, and outputs an analyzed result 16. Thus, the performance of the vector type related data base managing system can be evaluated, and the effect of hardware can be previously predicted.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to performance evaluation of relational database management systems using vector instructions.

[Conventional technology]

A conventional relational database management system is described in Computer Today magazine 198
515 NQ7F Relational System Question Processing System'', processing was performed for each record.

The main performance evaluation items for a relational database management system based on such a processing system were CI) 0 time and prediction of the number of inputs and outputs to the database. However, as a new processing system for relational database management systems, there is a vector type system in which records to be processed are grouped together and processed all at once by hardware.

[Problem that the invention seeks to solve]

In addition to the performance evaluation items of conventional relational database management systems, the application rate of vector instructions and the effectiveness of hardware become important items for evaluating the performance of vector-type relational database management systems. This new performance evaluation item cannot be achieved using conventional performance evaluation techniques.

An object of the present invention is to provide a performance evaluation method for a relational database management system that predicts in advance the effects of hardware in a vector-type relational database management system.

[Means for solving problems]

The present invention provides a method for predicting the effect of hardware in a vector-type relational database management system in advance.
When the relational database management system issues a request to execute a vector instruction, a vector instruction processing program that performs processing equivalent to hardware is executed, and then vector instruction execution information is recorded in a trace file via the trace area in main memory. However, after the processing of the relational database management system is completed, vector instruction execution information, which is the content of the trace file, is used to evaluate the performance of vector instructions, which is a direct effect of the hardware.

[Effect]

A relational database management system performs a data load function using trace files as input.

This creates a trace table Tz in the database. A hardware table T2 has already been created in the database. The relational database management system analyzes the analysis command input from the terminal using the trace table T1 and the hardware table T2, and outputs the analysis result.

[Embodiments of the invention]

An embodiment of the present invention will be described below with reference to the drawings.

In addition to relational database management systems, programs that use vector instructions include numerical calculations. Numerical calculation programs have a fixed processing procedure, and once the input data is determined, the length of the vector is also determined. For example, in the case of matrix calculation, once the size of the matrix is determined, the processing procedure is constant and the length of the vector is known, so performance can be predicted.

However, in a relational database management system, even if the input database operation command is determined, the processing procedure is optimized to speed up the processing, and the processing procedure cannot be predicted. Also.

The length of the vector cannot be predicted because it changes depending on the size of the table, the conditions of the database operation command, etc. Therefore, performance evaluation of a relational database management system requires execution information indicating in what environment each vector instruction was executed. Furthermore, since we want to know the performance before adding hardware, it is also necessary to create software that has the same functionality as the hardware.

1 and 2 show overall block diagrams of embodiments of the present invention. FIG. 1 is a diagram showing how vector instruction execution information is collected, and FIG. 2 is a diagram showing an embodiment in which performance evaluation is performed using the collected vector instruction execution information.

First, the constituent elements of the block will be explained according to FIG. 1 indicates the center, and the processing unit 2. Contains files 4 and 5 and main memory 1o. The file device 4 is a database used by a relational database management system that is subject to performance evaluation.

Reference numeral 5 is a file that stores execution information of vector instructions, and this is called a trace file. 3 is a terminal device into which commands for starting and stopping the relational database management system and commands for operating the relational database are inputted. The terminal 3 and files 4 and 5 are connected to the processing section 2. 1o is a main memory, and 7 contained therein is a group of main program modules of the relational database management system, and vector instructions are called from this main program module. 8 is a group of program modules that perform operations equivalent to vector instructions that should originally be executed by hardware. 6 is a program that has the function of controlling and managing program execution. This program also has means to transfer and set the contents of any area on the main memory to the file device. In this embodiment, the programs to be controlled and managed are program module groups 7 and 8. Reference numeral 9 denotes an area for storing execution information of vector instructions, and this area is called a trace area.

Here, the configuration of vector instruction execution information will be explained with reference to FIG. As shown in the figure, in this embodiment, execution information of seven types of vector instructions 81 to S7 was collected. Sl indicates the name of the executed vector instruction. The executed vector instruction can be determined by Sl. S2 indicates the name of the main program module that called the vector instruction. S3 indicates the name of the main program module that called Sz, and Sl indicates the name of the main program that called S3. From St~54, you can know the process by which a vector instruction is called. @ S8 and 88 are vector instructions Indicates the length of the input vector.

S7 indicates the length of the vector resulting from executing the vector instruction. 86 to S7 are elements that determine the execution time of the vector instruction.

Next, the operation shown in FIG. 1 will be explained. Control for inputting operation commands for the database for performance evaluation from the terminal 3 is transferred from the processing section 2 that receives the operation commands to the control/management program 6. After that, any program in the main program module group 7 of the relational database management system is executed.
The module starts running. P 1 "P- in the main program module 7 indicates the name of the program module. In this embodiment, Pm starts execution, calls Pz in the middle, and calls P'l in the middle of Pz. It was decided to call the vector instruction in the middle of the call and Pl.In the program module 8 that executes the vector instruction, v1 to vn indicate the name of the program module, and aRx to R, simulate the vector instruction. 81 to S0 are parts having the function of setting execution information of vector instructions to the trace area 9.The name Vz of the program module derived from the name PL of the main program module 7 is the vector Treatment i!
l After performing R1, the execution information is set in the trace area 9 by 81. Further, Sl uses the control management program 6 to transfer and set the contents of the trace area 9 to the trace file 5, and then returns. At this time, the execution information of the vector instruction is as shown in FIG.
1 g S x is P t g S a is Pz.

Sl is Pn, Sa and S8 are the lengths of the input vectors,
In step S7, an operation of 0 or more in which the length of the vector after execution of the vector instruction is stored is performed every time a vector instruction is executed, and the execution information of the vector instruction is accumulated in the trace file.

FIG. 2 is a diagram showing an embodiment in which performance evaluation is performed using the vector instruction execution information collected in FIG. 1.

In this embodiment, the application rate of vector instructions and the effectiveness of hardware were determined as items for evaluating the performance of the relational database management system. All you need to know is the number of times a vector instruction is used. Furthermore, the effectiveness of hardware can be determined from the execution time of the hardware and the time required for all processing.

The planting element shown in FIG. 2 will be explained. 5 is the same as in FIG. 1 and indicates a trace file.

Reference numeral 11 denotes a relational database management system, and here, a data load function, which is a function of the relational database management system, is used. The data load function stores the contents of an external file into a database. 13 indicates a database. 12 is the same relational database management system as 11, but here it is used as a means for performance analysis. 14 is a terminal.

This is for inputting analysis commands to the No. 12 relational database management system. 16 is the analysis result.

T1 and T2 in the database 13 indicate tables. Tz is a table created by executing the data load function of the relational database management system 11 using the trace file 5 as input, and is called a trace table. Contains information about the hardware it runs on.

Table Tz (hereinafter referred to as hard sea urchin table) will be explained with reference to FIG. Ez to Eδ indicate column names. Ex is the name of a vector instruction. Ea and E
δ and Ea are coefficients for determining the number of machine cycles required for execution from the length of the vector input to the vector instruction and the length of the output vector.

Here, a machine cycle is the basic time for an arithmetic unit to process data in a register in a processing device and store the result in a register.

The execution time of the command can be determined by (machine cycle required for a smooth hit) L41) x (execution time of one machine cycle). E8
is a value for correcting the number of machine cycles determined from the coefficients E2 to E4. Therefore, the calculation formula for determining the number of machine cycles is as follows.

Number of machine cycles for vector instruction name VNi S V N
aSVNJ = NIJXC input vector length 1) + N, L
XC input vector length 2) + N self × (output vector length %)
+NP J again. The execution time required for all the processing necessary for performance analysis can be known from the billing information that the center obtains when the computer is executed. From execution time to machine
To find the number of cycles, you can use (execution time)/(execution time of one machine cycle).

Next, the operation shown in FIG. 2 will be explained. The relational database management system 11 executes a data load function using the trace file 5 as input. As a result, a trace table T1 is created in the database 13. In the database 13, a hardware table T2 has been created. The relational database management system 12 analyzes the analysis command input from the terminal 14 using the trace table Tl and the hardware table T2, and outputs the analysis result 16. An example from when an analysis command is input from the terminal 14 to when the relational database management system 12 outputs an analysis result will be described below.

FIG. 5 is a diagram showing an example of performance analysis.

Tz and T2 are the same tables in the database 13 of FIG. 2, and are a trace table and a hardware table, respectively. The meanings of columns Sz to S7 of trace table T1 are the same as in FIG. Column W is used as a work column during analysis. The meanings of columns Ex to E11 of hardware table T2 are the same as in FIG. Tables T1 and T2 are used for analysis. When analysis command 1 is executed, the number of machine cycles executed by vector instruction RN- is set in software table Tl column W. 1 equation W = S x X E
2+5IIXE4+E6 is the same as 5VN- previously described. 1ilHEREΔS s = E t is a condition for extracting the coefficients necessary for calculating the expression from the hardware table T2. When analysis command 2 is executed, the vector instruction name, the number of times each vector instruction appears (COUNT (*)), and the total number of executed machine cycles (SUM (W)) are output (T3). From T8,
It can be seen that if the value of SUM(W) is a large proportion of the total number of machine cycles processed, it is better to improve the vector instruction. Furthermore, if we know the rate at which the vector instructions are improved, we can understand the effectiveness of the hardware, and at the same time, we can know the rate at which the processing speed of the relational database management system increases.

By applying the above method to performance analysis of a vector-type relational database management system, which was not possible using conventional performance evaluation techniques, it is possible to predict the effects of hardware in advance. Further, by using a relational database management system as a means for performing performance evaluation as in this embodiment, performance evaluation can be performed using relational database operation commands, and a wide range of performance analysis becomes possible. In addition, by taking advantage of the fact that the collection of vector instruction execution information and performance evaluation are independent, we have created a number of files that pair relational database operation commands that are subject to performance analysis and their trace results. Create database operation commands.

By doing so, even database users who cannot collect execution information can perform performance analysis, and many users can be given the opportunity to evaluate.

〔Effect of the invention〕

According to the present invention, by having means for outputting execution information of vector instructions to a file and means for performing performance evaluation using the contents of the file, a vector-type relational database management system that was not possible with conventional technology is provided. It is now possible to evaluate the performance of hardware and predict the effects of hardware in advance.

[Brief explanation of drawings]

Figure 1 is an explanatory diagram of collection of vector instruction execution information, Figure 2
The figure is an explanatory diagram showing an example of performance evaluation using collected vector instruction execution information, Figure 3 is a configuration diagram of vector instruction execution information, and Figure 4 is the hardware used during performance evaluation. FIG. 5 is a diagram showing an example of performance evaluation. 5... File device, 7, 8... Program module, 9... Vector instruction execution information storage area, 1
1.12...Relational database management system, 13.
...Database. Agent: Patent attorney Katsuma Ogawa. Figure 5 # UPDATE tsl”+ a 5ETa W
= 5zXEzfSsxEs+ 54XE+tf-5W
HEREASr =lE+ j44sett13-j2j-5r-ticrhr=1. :
avNr<t), sus<wtsua 4 FRDM + r
D2/dbuphgYbsr

Claims (1)

[Claims] In a processing device having a main memory section and a file device, and a relational database management system operating thereunder, a trace area provided in the main memory section that stores execution information of vector instructions, and a trace area provided in the main memory section that stores execution information of vector instructions; a trace file provided in the file device for storing;
means for writing execution information of vector instructions into the trace area; means for transferring and setting the contents of the trace area to the trace file; and means for evaluating the performance of the relational database using the contents of the trace file. In preparation, execution information of vector instructions executed during operation of the relational database management system is written to the trace area, the contents of the trace area are then transferred and set to the trace file, and after the processing of the relational database management system is completed, the A performance evaluation method for a relational database management system characterized by evaluating the performance of the relational database management system using the contents of a trace file.