JP3703549B2 - Software performance estimation and performance monitoring method - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a software performance estimation and performance monitoring method.
[0002]
[Prior art]
In recent years, there has been a demand for computer systems to increase the amount of data or transactions that can be processed. And even when the system scale increases, it is required to extract the maximum performance of the system. Conventionally, performance estimation at the time of system design has been indispensable in order to extract the maximum performance of the system. In addition, after starting the business, it was necessary to continuously monitor the system performance to see if the target performance was as designed.
[0003]
Here, in performance estimation at the time of design, an estimation formula is used in which the number of resources that increase according to the system scale becomes a variable. For this reason, particularly in a large-scale system, the substitution value to the estimation formula is determined considering the safety of system operation as a first priority. For example, the maximum load is set to a maximum value, the maximum value is multiplied by a safety factor to give a margin, or a large value is set in anticipation of future system expansion.
[0004]
For example, if the substitution value is a variable called the number of files, and if the numerical value including files temporarily created for work is 100, a numerical value of 120 multiplied by a safety factor of 1.2 is set. Alternatively, if the substitution value is a variable called the number of logged-in users, even if the maximum number of logged-in users at the beginning of business is limited to 300, if an increase in demand is expected in two to three years ahead, A large value of 400 is set in advance.
[0005]
As described above, the performance estimation often assumes an environment in which a process to be estimated operates or a system load at that time in the worst case. In such a conventional performance estimation method, various problems occur when the estimation based on the assumed load exceeds the allowable range in the design of the system. For example, even if the estimate based on the assumed load exceeds the allowable range in the design of the system, when the system is operating at a low load, such as the number of logged-in users is only 20th in the early morning, Processing can be performed. However, the conventional method for estimating the performance requires such measures as making the processing completely unusable or limiting the number of concurrent executions.
[0006]
As a specific example of this correspondence, performance estimation in the performance monitoring monitor will be described. The performance monitoring monitor displays the performance status of the system in real time .
As a result of SE's performance estimation for such a performance monitoring monitor, a function for displaying the performance status of a particularly large number of resources connected / configured to the system is used in daytime online operations. Suppose that there is an influence. In this case, on the user side, two methods are conceivable: a method that does not use only the function that has an influence on the operation side, or a method that corrects the program to disable the function.
[0007]
Here, if it is judged that guarding from the operation side is impossible because the operator may make an operation mistake, the function is not only included in daytime online work but also in nighttime batch work. A program fix is made to make it completely unusable. Specifically, the following two corrections are made.
a. Fix not to accept request by outputting error message even if PF key to call the function is pressed.
[0008]
b. A fix that prevents more than the specified number of performance monitoring monitors from starting.
[0009]
[Problems to be solved by the invention]
As described above, in the performance estimation at the time of conventional system design, the situation when the system load reaches a peak is taken into consideration. For this reason, even if the processing can be used when the system load is light, problems such as making it completely unusable or adding a program correction for it have arisen. In addition, there is a problem that it is difficult to obtain a highly reliable estimate because the estimate is based on an assumed load or a gap with an actual measurement value occurs after the system is operated due to a calculation error or the like.
[0010]
If a program that exceeds the system design tolerance is run continuously at a high priority due to an inadequate estimate, control will not be transferred to other programs, causing various problems. There was also.
It is an object of the present invention to provide a method that eliminates the need for performance estimation at the time of system design and can automatically perform software performance estimation and performance monitoring during system operation.
[0011]
[Means for Solving the Problems]
The present invention is a software performance estimation and performance monitoring method for achieving the above-described object, and immediately before the start of execution of each process of the software, the estimation formula built in the program of the process and the variable at that time are calculated. To calculate the processing load . This load is compared with a reference value, for example, a predetermined value set in advance or a load state of the system. Based on the result, it is determined whether or not the process can be executed.
[0012]
As a result of the determination, if it is determined that the process can be executed, the process is executed. If it is determined that the process cannot be executed, the control of the process, for example, the abandonment of the process, the waiting for the execution of the process, the adjustment of the priority of the execution of the process, or every fixed period or every fixed resource use Execution control by.
According to the present invention, it is possible to automatically estimate an accurate overhead in real time during system operation without performing performance estimation during system design. Further, it becomes possible to execute processing or control processing according to the system load without newly modifying the program. Therefore, the reliability of the computer system can be improved by the present invention.
[0013]
DETAILED DESCRIPTION OF THE INVENTION
An embodiment of the present invention will be described with reference to FIG.
In the figure, reference numeral 11 denotes one of the processes included in the software. This process means a unit such as a function, a subroutine, a function, or a job. The illustrated process 11 includes a resource collection unit 1, an estimation formula execution unit 2, an external recording unit 3, a determination unit 4, and a control unit for executing performance estimation and performance monitoring before the processing body 6 to be estimated. 5 is built-in. Note that the resource collection unit 1 to the control unit 5 can be called from other sources without being included in the processing. In the following description, the resource collection unit 1 to the control unit 5 are collectively referred to as “main mechanism”.
[0014]
First, the operation of this mechanism shown in FIG. 1 will be described.
The resource collection unit 1 collects the number of resources, and collects values to be substituted into the variables of the estimation formula. If the assigned value relates to a resource handled in other processing, a resource collection subroutine is called, or a resource number notification interface is provided by referring to the value assigned to the OS control table. On the other hand, if the assigned value is a resource handled by the own process 11, the own process itself knows it, and thus no new interface is required.
[0015]
The estimation formula execution unit 2 executes the estimation formula by substituting the values collected by the resource collection unit 1 into the estimation formula variable. This estimation formula is built in at the head position of the program of its own processing. As an example of the estimation formula, if a certain process is shown in FIG. 2, the number of instructions is represented by an estimation formula of “23+ (23+ (45 + number of extents × 3)) × number of extent types + 5126”. The Here, when the substitution values of the number of extents and the number of types of extents are 33 and 9, the load of a certain process can be calculated as 6,652 steps. In the following description, the estimation formula is abbreviated as “estimation formula A”.
[0016]
Returning to FIG. 1, the external recording unit 3 records the estimation result of each processing 11 in an external file or displays it on a display device. If the estimation result of each process is recorded in an external file or the like, it can be verified whether the estimation at the time of design is correct by comparing the estimate at the time of system design with the recorded estimation result. Here, when the target performance determined by the system department is not achieved, it is possible to investigate what the processing with different loads is, and the cause can be easily determined.
[0017]
The determination unit 4 determines whether or not to execute the process. The determination unit 4 compares the estimation result of the estimation formula execution unit 2 with a reference value. Based on the comparison result, it is determined whether the transition is made to the control unit 5 (the process is not executed) or the control unit 5 is skipped (the process is executed). There are the following two examples of comparison. Which method is used for comparison can be designated from the outside for each process.
[0018]
(1) Comparison with user-defined reference value The estimated value is compared with a reference value (predetermined value) defined by the user in a file or the like. For example, if the reference value is defined as 6,000 steps in Example A above, it is determined that the process is not executed because the estimate is large by 652 steps.
Alternatively, a definition with a range such as 6,000 + 10% is allowed as a reference value, execution is permitted if the estimated value is within 6,000, and execution wait or priority is in the range of 6,001 to 6,600. If it exceeds 6,600, it can be determined not to execute. Note that the user-defined reference value is not limited to a dynamic step, and may be converted into a unit of CPU time or CPU load factor.
[0019]
(2) Comparison with system load status The system load status at the time of estimation is compared with the estimation result. Taking the CPU load as an example of the system load status, a comparison with a system margin value (maximum capacity-actual measurement value) as shown in FIG. 3 can be considered. This actual measurement value can be obtained by a real-time API. The result of actual measurement by the real-time API is updated asynchronously with the own process at regular intervals in an area where each process can be referred. In the case of the above estimation formula A, the estimated value is converted into the same unit as the actual measurement value by the following formula prior to the comparison.
[0020]
Estimated value [%] = (average instruction execution time per instruction × number of dynamic steps / real-time API update period) × 100
If the average instruction execution time per instruction is 10 mS and the update period of the real-time API is 1 minute, 1,652 steps are 10 mS × 1652 ÷ 60000 mS × 100 = 27.5%. When the actual measurement value of 80% is obtained, the margin value is 20%, and since the estimated value 27.5% is larger than the margin value 20%, it can be determined that the process is not executed. Of course, the margin value may have a width as in the case of the reference value (1).
[0021]
In the estimated value equation, the dynamic step in the case of a high-level language (COBOL, C) can also be obtained by the average number of expanded instructions per instruction of language × the average instruction execution time per instruction.
Returning to FIG. 1, the control unit 5 controls the process when the process is not executed as a result of the determination by the determination unit 4. There are the following four examples of the control method. Which method is to be controlled can be designated from the outside for each process.
[0022]
(1) Set a return code indicating processing abandonment and return to the caller of the processing. As a matter of course, the interface with the caller must be protected so that no inconvenience arises because the processing is not performed.
(2) Waiting Enters the waiting for a fixed time, and after that time, returns to the resource collection unit 1 again. The amount of time to wait depends on the nature or attributes of the process, but can be instructed from the outside.
[0023]
(3) Execution priority adjustment After the execution priority is adjusted (generally, the priority is lowered), the return of (1) is performed. How much priority is adjusted may be designated from the outside, or may be left to the system.
(4) Control by fixed period or fixed resource use Generates an interrupt after a fixed time elapses, or generates an interrupt for each fixed resource use such as CPU time, and executes the process while interrupting the process at each fixed period To do. (Techniques that generate interrupts at fixed intervals and at constant resources, or resume processing interrupted by interrupts are known)
Although a bypass route is shown in FIG. 1, this is used when the estimation accuracy at the time of system design is high and this mechanism is useless.
[0024]
As described above, according to this example, performance estimation and performance monitoring can be executed simultaneously. This eliminates the need for performance estimation during system design. According to this example, an accurate overhead corresponding to the system load can be automatically re-estimated in real time during system operation. It is also possible to determine whether the execution of the estimation target process has an effect on the system operation, and the execution of the process can be controlled.
[0025]
In the above, the estimation of the CPU overhead has been described as an example. However, for example, the concept is the same for memory or disk resources, and performance estimation and performance monitoring are performed by changing the estimation formula and variables in the same way. Can do.
Next, an example in which the above-described performance estimation and performance monitoring method is applied to various processes will be described.
[0026]
FIG. 4 shows an example applied to the performance monitoring monitor.
The performance monitoring monitor 42 is connected between the real-time API of the host computer 41 and the personal computer 43. Since these structures are well-known, detailed description is abbreviate | omitted.
The mechanisms 44 and 45 shown in FIG. 4 are incorporated in the response monitoring / processing time / processing count processing, buffer shortage resource detection processing, and the like of the performance monitoring monitor 42. If the estimation result of response / processing time / number of processes, detection of resource shortage of buffer, etc. exceeds the reference value, the processing of the performance data is not executed and the performance data is not transmitted to the personal computer 43.
[0027]
Therefore, the data transmitted to the personal computer 43 is the performance data of the constant monitoring screen, the performance data below the reference value for other processing, and an error message, which are displayed on the sub screen of the personal computer 43.
FIG. 5 shows an example of application to a compiler.
The compiler outputs an object 51 containing the mechanism 52 and estimation formulas 53, 54,. The estimation formulas 53 and 54 of each function / subroutine are called recursively.
[0028]
If the application does not meet the performance target, the execution of the application can be abandoned by itself.
When applied to a debugger, in addition to breakpoints due to instruction offsets, breakpoints based on performance values (such as a break when a subroutine whose performance values are not satisfied) can be specified.
[0029]
Furthermore, the program can be suspended or resumed from the viewpoint of performance monitoring during application execution. This is effective in application testing, particularly in an actual environment where the application is operated. Based on the log of the estimation result, the program can be improved and the estimated value can be verified.
[0030]
In addition, as a measure when the standard value is not satisfied during application execution, it is possible to specify whether to suspend the process on the spot or continue with the compile option based on the comparison with the system status. If it is busy, it can also be used to control execution, such as waiting for application execution.
[0031]
FIG. 6 shows an application example to the download process.
When downloading data via PC communication, you will be charged for the time, so you will be wondering how long it will take to download. In order to cope with this, as shown in FIG. 6, the main mechanism 63 is connected between the downloader 62 of the personal computer communication center 61 and the personal computer 64.
[0032]
This mechanism 63 obtains the required time from the connected line speed and the number of transfer blocks. For example, a value of 10 minutes is designated as the reference value. If it takes more time as a result of the estimation of the mechanism 63, the download can be automatically stopped. Further, by comparing with the line load at that time, it is possible to automatically switch to an available line or to wait for a while until the load decreases.
[0033]
An example of application to a copy command will be described.
Personal computer software is restricted by CPU speed, memory capacity, and disk capacity, and the software may not operate depending on the configuration. Therefore, it is necessary to conduct a preliminary survey before purchasing software.
In order to cope with this, when the command for copying from the software providing medium to the personal computer copies the program / data, it passes control to the estimation formula of the program to be copied, and performs estimation based on the personal computer environment. If it is within the reference value, copying is permitted, otherwise copying is not permitted.
[0034]
According to this method, even if the copy destination system does not actually exist, by preparing the basic data for the estimation assuming the copy destination system, it is possible to input the provided medium and which computer can be used. You can check it without actually using it on the computer.
[0035]
【The invention's effect】
According to the present invention, performance estimation at the time of system design is not required, and software performance estimation and performance monitoring can be automatically performed in real time during system operation. This makes it possible to control the execution of processing according to the system load without newly modifying the program. According to the present invention, the reliability of a computer system can be improved.
[Brief description of the drawings]
FIG. 1 shows a mechanism for carrying out the method of the present invention.
FIG. 2 is a diagram showing an example of an estimation formula used in the present invention.
FIG. 3 is a diagram for explaining a system load situation of a reference value used in the present invention.
FIG. 4 is a diagram showing an example in which the present invention is applied to a performance monitoring monitor.
FIG. 5 is a diagram showing an example in which the present invention is applied to a compiler.
FIG. 6 is a diagram showing an example in which the present invention is applied to download processing.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 ... Resource collection part 2 ... Estimation formula execution part 3 ... External recording part 4 ... Determination part 5 ... Control part 6 ... Processing main body

Claims (7)

Using a computer, immediately before the start of execution of software processing, the load of the processing program is calculated using an estimation formula built in each processing program , and the value is a reference value prepared for each processing program. A performance estimation and performance monitoring method for software, which determines whether to execute the process or to control the process by comparing with, and executes or controls according to the determination.   2. The software performance estimation and performance monitoring method according to claim 1, wherein the reference value is a predetermined value.   2. The software performance estimation and performance monitoring method according to claim 1, wherein the reference value is a CPU load.   4. The software performance estimation and performance monitoring method according to claim 1, wherein the control of the process is abandonment of the process.   The software performance estimation and performance monitoring method according to claim 1, wherein the control of the process is waiting for execution of the process.   4. The software performance estimation and performance monitoring method according to claim 1, wherein the control of the process is to adjust a priority of execution of the process. 5.   The software performance estimation and performance monitoring method according to any one of claims 1 to 3, wherein the control of the processing is performed at a constant cycle or at every constant resource use.

Images