CN107451022B - A kind of method and system automatically adjusting linpack performance test - Google Patents

Abstract

It includes a method for automatically adjusting the linpack performance test, which specifically includes the following steps: calculating the theoretical value and expected value of the CPU performance; continuously adjusting the N value and the NB value to realize the automatic adjustment of the linpack performance test. It also includes a system for automatically adjusting the linpack performance test, including a CPU performance value calculation module, which is used to read the number of cores, the number of CPU cores and the main frequency information, and calculate the theoretical value and expected value of the CPU performance; the automatic tuning module, Used to modify the NB value and N value of the configuration file, N represents the operation order of linpack, and NB represents the size of the solution matrix block. It can realize automatic tuning of linpack performance test, realize automatic configuration, automatically run tests, output optimal results, and do not need to manually modify configuration files, save human resources and testing time resources, and realize simplification and automation of testing.

Description

A method and system for automatically adjusting linpack performance test

technical field

The invention relates to the technical field of performance testing, in particular to a method and system for automatically adjusting linpack performance testing.

Background technique

Linpack (linear system software package) has now become the most popular benchmark (benchmark) for testing the floating-point performance of high-performance computer systems in the world.

With the continuous upgrading of server product hardware, the entire computing power has also increased by an order of magnitude. An important indicator to measure computer performance is the calculation peak, such as the floating-point calculation peak, which refers to the maximum number of floating-point calculations that the computer can complete per second. The measured floating-point peak value refers to the Linpack test value, that is to say, the Linpack test program is run on this machine, and the optimal test results are obtained through various tuning methods. The most common tuning method is to modify the linpack configuration file. to tune.

At present, most of the modification configuration files are manually modified, which is not only time-consuming but also prone to errors.

SUMMARY OF THE INVENTION

In order to solve the problems raised in the background art, the present invention provides a method and system for automatically adjusting the linpack performance test.

The technical scheme adopted by the present invention to solve the technical problem is: a method for automatically adjusting linpack performance test, which specifically includes the following steps:

Calculate the theoretical value and expected value of CPU performance;

Continuously adjust the N value and NB value to realize automatic adjustment of linpack performance test.

Further, before the above steps, it is necessary to obtain the path of linpack, and call the /proc/cpuinfo file under the redhat system to obtain the number of cores, the number of CPUs, and the main frequency of the CPU.

Further, the calculation method of the theoretical value of the CPU performance is: theoretical value = the number of CPUs * the number of cores of the CPU * the main frequency * the number of times that the CPU performs floating-point operations per clock cycle;

The calculation method of the expected value of CPU performance is: expected value=theoretical value*90%.

Further, the method for realizing automatic adjustment specifically includes the following steps:

Set the initial value of N value, and perform performance test on linpack to determine the optimal value of NB;

According to the optimal value of NB, the N value is continuously adjusted to make linpack achieve the best performance.

Further, the method for continuously adjusting the value of N specifically includes the following steps:

Set the initial value of error and derror(t)/dt;

Substitute error and derror(t)/dt into the formula N1=N+Kp*error+KD*derror(t)/dt to obtain a new N value: N1; in the formula, KD is the rate of change of the difference, and Kp is the control The proportional part of the output of the device is proportional to the error, and error is the error;

Use N1 to test linpack and get the test result value;

Let error = test result value - expected value, judge whether the error is greater than 0, if it is greater than 0, output the result; otherwise, perform the next operation;

Substitute the new error value into the formula N1=N+Kp*error+KD*derror(t)/dt, get the new N value again: N2, continue to use N2 to test the linpack, until the new error value is continuously measured Greater than 0.

A system for automatically adjusting a linpack performance test, using the method, comprising a CPU performance value calculation module for reading information on the number of cores, the number of cores and the main frequency of the CPU, and calculating the theoretical value and expected value of the CPU performance; and ,

The automatic tuning module is used to modify the NB and N values of the configuration file. N represents the operation order of linpack, and NB represents the size of the solution matrix block.

Further, it also includes a file path selection module, which is used to obtain the absolute path of the linpack software, and returns the path to the background as a variable, and calls the /proc/cpuinfo file under the redhat system to obtain the number of cores and chips of the CPU. , the dominant frequency; and,

The result output module is used to output the test result of the lin pack to the display device.

The effects provided in the summary of the invention are only the effects of the embodiments, rather than all the effects of the invention. One of the above technical solutions has the following advantages or beneficial effects:

It can realize automatic tuning of linpack performance test, realize automatic configuration, automatically run tests, output optimal results, and do not need to manually modify configuration files, save human resources and testing time resources, and realize simplification and automation of testing.

Description of drawings

1 is a schematic flowchart of a method according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of a system connection according to an embodiment of the present invention.

Detailed ways

In order to clearly illustrate the technical features of the solution, the present invention will be described in detail below through specific embodiments and in conjunction with the accompanying drawings. The following disclosure provides many different embodiments or examples for implementing different structures of the invention. In order to simplify the disclosure of the present invention, the components and arrangements of specific examples are described below. Furthermore, the present invention may repeat reference numerals and/or letters in different instances. This repetition is for the purpose of simplicity and clarity and does not in itself indicate a relationship between the various embodiments and/or arrangements discussed. It should be noted that the components illustrated in the figures are not necessarily drawn to scale. Descriptions of well-known components and processing techniques and processes are omitted from the present invention to avoid unnecessarily limiting the present invention.

As shown in Figure 1, a method for automatically adjusting the linpack performance test includes the following steps:

Step 1) Calculate the theoretical value and expected value of CPU performance;

Step 2) Continuously adjust the N value and the NB value to realize automatic adjustment of the linpack performance test.

Before the above steps, you also need to obtain the path of linpack, and call the /proc/cpuinfo file under the redhat system to obtain the number of cores, the number of CPUs, and the main frequency. It is necessary to obtain the absolute path of the linpack software through the path selection module for the user to test.

The calculation method of the theoretical value of the CPU performance is: theoretical value = the number of CPUs * the number of cores of the CPU * the main frequency * the number of times the CPU performs floating-point operations per clock cycle; if it is used to test the CPU of the purely platform, each CPU The number of floating-point operations performed per clock cycle is a constant 32, and the number of floating-point operations performed by the CPU per clock cycle is platform-dependent.

The calculation method of the expected value of CPU performance is: expected value=theoretical value*90%. The server test pass standard result is about 90% of the theoretical value, so the expected result of the setup test is 90% of the theoretical value.

The method for realizing automatic adjustment specifically includes the following steps:

Step 21) Set the initial value of the N value, and perform a performance test on the linpack to determine the optimal value of NB;

Step 22) According to the optimal value of NB, continuously adjust the value of N, so that the performance of linpack is optimal.

The method of continuously adjusting the value of N includes the following steps:

Step 221) Set the initial values of error and derror(t)/dt; the initial value of error is 1 by default, and the initial value of Derror(t)/dt is 0 by default.

Step 222) Substitute error and derror(t)/dt into the formula N1=N+Kp*error+KD*derror(t)/dt to obtain a new N value: N1; in the formula, KD is the rate of change of the difference, Kp is the proportional part of the controller output and the error is proportional, error is the error;

Step 223) Test linpack with N1, and obtain the test result value; after calculating the new N value, it will automatically enter the directory where the linpack configuration file HPL.dat is located, modify the HPL.dat configuration file, and modify the value of parameter N In order to calculate the new N value, after modifying and saving, call the execution file of linpack for testing. The result of the test is redirected and saved in the /home directory. After the test is completed, linpack will output end of test, and the script judges the test completion according to the string. , enter the /home directory to read the test result file, filter the content of the file, first find the word Gflops in the file to locate, and get the last number of the next line as the result

Step 224) make error=test result value-expected value, judge whether error is greater than 0, if greater than 0, then output the result; otherwise, execute the next step;

Step 225) Substitute the new error value into the formula N1=N+Kp*error+KD*derror(t)/dt, and obtain the new N value again: N2, continue to use N2 to test the linpack, until the new N2 is continuously measured. The error value is greater than 0.

N is the operation order of linpack. The larger N is, the larger the proportion of effective calculation is, and the higher the system floating-point processing performance is. However, the increase of N will lead to an increase in memory consumption. Once the memory space is insufficient, using the cache will reduce the performance drastically reduced.

NB is to solve the size of the matrix block. In order to improve the locality of the data and thus improve the overall performance, the size of the block has a great impact on the performance. NB*8 must be a multiple of Cacheline. Through empirical tests, NB is generally It is 384, 192, 256 three groups of values, the software sets NB as an array [384, 192, 256].

The values of KD and KP can be queried according to the experience table; for example, the value of KD can be queried from Table 1:

rate of change\error <100 100-500 500-1800 1800-2500 2500> <10 PB PB PB z z 10-20 PB PS PS z z 20-25 PB PS z NS NB 25-30 z z NS NS NB 30> z z NB NB NB

Table 1

NB, PB, Z, and NS in Table 1 are specific numerical values, and are replaced by NB, PB, Z, and NS here.

The value of KP can be queried from Table 2:

Table 2

NB, PB, Z, and NS in Table 2 are specific values, and are replaced by NB, PB, Z, and NS here.

As shown in Figure 2, a system for automatically adjusting the performance test of linpack, including a CPU performance value calculation module, which is used to read the number of cores, the number of CPU cores and the main frequency information, and calculate the theoretical value and expected value of the CPU performance; and ,

The automatic tuning module is used to modify the NB value and N value of the configuration file, where N represents the operation order of linpack, and NB represents the size of the solution matrix block; and,

The file path selection module is used to obtain the absolute path of the linpack software, and returns the path to the background as a variable, and calls the /proc/cpuinfo file under the redhat system to obtain the number of cores, number of CPUs, and main frequency; and ,

The result output module is used to output the test result of the lin pack to the display device.

The above are only the preferred embodiments of the present invention. For those skilled in the art, without departing from the principles of the present invention, several improvements and modifications can be made, and these improvements and modifications are also regarded as the present invention. The scope of protection of the invention.

Claims (5)

1. a method for automatically adjusting linpack performance test, is characterized in that, specifically comprises the following steps: Calculate the theoretical value and expected value of CPU performance; Continuously adjust the N value and NB value to realize automatic adjustment of linpack performance test; The calculation method of the theoretical value of the CPU performance is: theoretical value=number of CPUs*number of cores of CPU*main frequency*number of times that the CPU performs floating-point operations per clock cycle; The calculation method of the expected value of the CPU performance is: expected value=theoretical value*90%; The method for continuously adjusting the N value specifically includes the following steps: Set the initial value of error and derror(t)/dt; Substitute error and derror(t)/dt into the formula N1=N+Kp*error+KD*derror(t)/dt to obtain a new N value: N1; in the formula, KD is the rate of change of the difference, and Kp is the control The proportional part of the output of the device is proportional to the error, and error is the error; Use N1 to test linpack and get the test result value; Let error = test result value - expected value, judge whether the error is greater than 0, if it is greater than 0, output the result; otherwise, perform the next operation; Substitute the new error value into the formula N1=N+Kp*error+KD*derror(t)/dt, get the new N value again: N2, continue to use N2 to test the linpack, until the new error value is continuously measured Greater than 0. 2. method according to claim 1, is characterized in that, also needs to obtain the path of linpack before above-mentioned steps, calls the /proc/cpuinfo file under redhat system, in order to obtain the number of cores, number of pieces, main frequency of CPU . 3. The method according to claim 1, wherein the method for realizing automatic adjustment specifically comprises the following steps: Set the initial value of N value, and perform performance test on linpack to determine the optimal value of NB; According to the optimal value of NB, the N value is continuously adjusted to make linpack achieve the best performance. 4. a system of automatic adjustment linpack performance test, utilizes the method described in any one of claims 1 to 3, is characterized in that, comprises CPU performance value calculation module, is used to read the number of cores, number of particles and main CPU of CPU. frequency information, and calculate theoretical and expected values of CPU performance; and, The automatic tuning module is used to modify the NB value and N value of the configuration file. N represents the operation order of linpack, and NB represents the size of the solution matrix block. 5. system according to claim 4, is characterized in that, also comprises file path selection module, is used to obtain the absolute path of linpack software, and path is returned to background with a variable, calls /proc/cpuinfo under redhat system file to obtain the number of cores, cores, and main frequency of the CPU; and, The result output module is used to output the test result of the lin pack to the display device.