CN103914121B - Multicomputer system and method and device for optimizing power consumption of same - Google Patents

Abstract

The present invention provides a multi-machine system, a method and a device for optimizing the power consumption of the multi-machine system, by determining the number of data processing devices used to adjust the power consumption of the multi-machine system within the determined number range of each search The first test point and the second test point are two test points, so that after each search, the interval on the side of the test point with a larger power consumption value will be discarded, and the test point will be used as the boundary of the next search core range, effectively The search range of the number of cores is greatly narrowed, and the power consumption optimization efficiency of the multi-computer system is improved.

Description

Multi-machine system, method and device for optimizing power consumption of multi-machine system

technical field

The invention relates to computer energy-saving technology, in particular to a multi-machine system, a method and a device for optimizing power consumption of the multi-machine system.

Background technique

Processor power management technology is an important topic in processor design in recent years. With the advancement of deep submicron technology, leakage power consumption has become an integral part of processor power consumption. Therefore, a series of technologies aimed at reducing processor leakage power consumption (static power consumption) have been born.

The first method widely used to reduce leakage power consumption of processors is Dynamic Power Management (DPM). DPM first reduces unnecessary power consumption overhead by shutting down idle processors or processor cores, and then reduces power consumption by migrating tasks and shutting down processors or processor cores with lower loads.

Secondly, with the widespread application of Dynamic Voltage and Frequency Scaling (DVFS), combining DVFS with the aforementioned DPM can improve the performance of other working cores while shutting down processors or processor cores with lower loads. Voltage frequency can save power consumption and ensure performance without loss.

However, under the premise of ensuring performance, it is not that the higher the frequency and the fewer the number of cores, the lower power consumption will be obtained.

On the one hand, the increase in frequency will lead to a super-linear increase in power consumption, so when the increase in dynamic power consumption caused by the increase in frequency is greater than the decrease in static power consumption caused by turning off the processor core, the total power consumption of the processor will increase; On the other hand, for programs with high parallelism, the performance can be guaranteed by increasing the number of execution cores and reducing the frequency, but when the increase in the number of cores brings about a static power consumption increase greater than the reduction in dynamic power consumption caused by frequency reduction, the processor's The total power consumption will also increase. Therefore, under the premise of ensuring the same performance, if the number of execution cores of the processor is adjusted while using DVFS, the power consumption will first decrease and then increase as the number of cores increases.

For the above rules, for the case of running large-scale multi-threaded programs on multi-core or even many-core processors, how to find the frequency of the optimal number of execution cores for different programs under certain performance constraints becomes the ultimate goal of power consumption optimization management.

Currently, turbo boost (Turbo boost) technology is a power management method used in Intel (Intel) mainstream processors. This technology adjusts the frequency of the processor core through the underlying hardware, and can operate the frequency of a specified single core, while the rest of the unloaded processor cores enter a deep sleep state to achieve a balance between power consumption and performance.

However, Turbo boost technology is mainly applied to processors with less than or equal to 8 cores, and the number of cores for mainstream processors is small. If the number of cores increases and the frequency of cores with high load is increased, it is very likely that the power consumption will first decrease and then increase with the increase of the number of cores. Moreover, the parallelism of the applications targeted by this technology is limited, resulting in the number of threads being usually smaller than the number of processor cores. At this time, turning off idle processor cores can reduce static power consumption, but when the number of program threads is greater than the number of processor cores, turning off Some processor cores may increase the load on other processor cores, making it impossible to guarantee target performance or resulting in increased power consumption.

Another power management method is to use the hill-climbing method to search for the number of cores required for the lowest power consumption in the power consumption-core space. This method takes a certain number of cores a as the test point, measures the power consumption value, and then executes it on a+1 processor cores. If the power consumption is greater than the power consumption value measured on a core, the next test point is a-1 cores; if the power consumption is less than the power consumption measured on a core, the number of cores at the next test point is a+2, and it is executed in turn on both sides of the core number a, and the corresponding power consumption, and find the number of processor cores corresponding to the lowest power consumption.

The disadvantage of this method is that the test point approaches the number of cores where the lowest power consumption is slow, and each test point increases or decreases by one core number compared to the previous one. As the scale of the processor increases, the number of trials of the hill-climbing search and traversal will also increase greatly, and the speed of obtaining the optimal solution is relatively slow. Therefore, the scalability of this method is relatively poor, and it cannot quickly approach the optimal value of power consumption.

Contents of the invention

Embodiments of the present invention provide a multi-machine system, a method and an apparatus for optimizing power consumption of the multi-machine system, and are used for improving power consumption optimization efficiency of the multi-machine system.

In the first aspect, an embodiment of the present invention provides a method for optimizing power consumption of a multi-machine system, including:

Determining the number range of data processing devices used to adjust the power consumption of the multi-machine system in the multi-machine system, the minimum value of the number range is the lower bound, and the maximum value is the upper bound;

Search within the number range to determine the first test point and the second test point; the first test point and the second test point are the number of data processing equipment, and the first test point and the second test point The sum of the two test points is equal to the sum of the upper bound and the lower bound;

Shut down all non-executing data processing devices according to the first test point and the second test point, and gradually reduce the frequency of the remaining data processing devices to meet the target performance.

In the second aspect, an embodiment of the present invention provides an apparatus for optimizing power consumption of a multi-machine system, including:

A range determination unit, configured to determine the range of the number of data processing devices used to adjust the power consumption of the multi-machine system in the multi-computer system, the minimum value of the number range is a lower bound, and the maximum value is an upper bound;

The test point determination unit is used to search within the quantity range to determine the first test point and the second test point; the first test point and the second test point are the numbers of data processing equipment, and the first test point The sum of a test point and said second test point is equal to the sum of said upper bound and said lower bound;

A performance adjustment unit, configured to shut down all non-executing data processing devices according to the first test point and the second test point, and gradually reduce the frequency of the remaining data processing devices to meet the target performance.

In a third aspect, an embodiment of the present invention provides a multi-machine system, including a multi-machine system body and the above-mentioned device for optimizing power consumption of the multi-machine system, and the device is used for optimizing power consumption of the multi-machine system body.

The multi-machine system, method and device for optimizing the power consumption of the multi-machine system provided by the embodiments of the present invention determine the power consumption of each search within the determined number of data processing devices used to adjust the power consumption of the multi-machine system The first test point and the second test point are two test points, so that after each search, the interval on the side of the test point with a larger power consumption value will be discarded, and the test point will be used as the boundary of the next search core range, The search range of the number of cores is effectively narrowed, and the power consumption optimization efficiency of the multi-computer system is improved.

Description of drawings

FIG. 1 is a flowchart of a method for optimizing power consumption of a multi-machine system provided by an embodiment of the present invention;

FIG. 2 is a flowchart of another method for optimizing power consumption of a multi-computer system provided by an embodiment of the present invention;

FIG. 3 is a schematic structural diagram of an apparatus for optimizing power consumption of a multi-computer system provided by an embodiment of the present invention;

FIG. 4 is a schematic structural diagram of a multi-machine system according to an embodiment of the present invention.

detailed description

FIG. 1 is a flowchart of a method for optimizing power consumption of a multi-computer system provided by an embodiment of the present invention. The method provided in this embodiment can be implemented by adding a device in the multi-computer system, for example, setting a function module in the multi-computer system. Or another device is provided to implement the method. As shown in Figure 1, the method includes:

Step 11: Determine the number range of the data processing devices used to adjust the power consumption of the multi-machine system in the multi-machine system, the minimum value of the number range is the lower limit, and the maximum value is the upper limit.

Wherein, the multi-computer system may be a multi-core processor, a system with multiple processors, or a system with multiple devices whose frequencies can be adjusted independently.

Step 12. Search within the above quantity range to determine the first test point and the second test point; the first test point and the second test point are the numbers of data processing equipment, and the first test point and the second test point The sum of the test points is equal to the sum of the upper bound and the lower bound.

Step 13. Shut down all non-executing data processing devices according to the first test point and the second test point, and gradually reduce the frequency of the remaining data processing devices to meet the target performance.

In this embodiment, two test points, the first test point and the second test point, are determined for each search within the determined number of data processing devices used to adjust the power consumption of the multi-computer system, so that each search Finally, the interval on the side of the test point with a large power consumption value will be discarded, and the test point will be used as the boundary of the next search for the number of cores, which effectively narrows the search range for the number of cores and improves the power consumption optimization of the multi-computer system. efficiency.

In the above step 11, determining the number range of data processing devices used to adjust the power consumption of the multi-machine system in the multi-machine system may include: determining the data processing equipment used in the multi-machine system to adjust the power consumption of the multi-machine system The initial quantity range of devices.

Specifically, determining the initial quantity range of the data processing equipment used to adjust the power consumption of the multi-computer system in the multi-computer system may include:

If the number of threads running in the multi-computer system is greater than the number of all data processing devices in the multi-computer system, the minimum value of the initial number range is 0, and the maximum value is all data processing devices in the multi-computer system quantity;

If the total number of threads or the number of running threads in the multi-computer system is less than the number of all data processing devices in the multi-computer system, the minimum value of the initial number range is 0, and the maximum value is 0 in the multi-computer system The number of threads running in .

In the above step 12, searching within the quantity range to determine the first test point and the second test point may include:

Calculate a=(X+Y)×m, b=(X+Y)×(1-m); where a is the first test point, b is the second test point, 0<m<1 , X is the upper bound, Y is the lower bound, and a, b, X, and Y are variables;

Execute all threads in the multi-computer system with the a data processing device, and measure the power consumption A of the multi-computer system;

Execute all threads in the multi-computer system with the b data processing devices, and measure the power consumption B of the multi-computer system;

compare said A and B;

If |A-B|<w, execute the step of shutting down all non-executing data processing devices according to the first test point and the second test point, where w is a first predetermined value;

If |A-B|>=w, judge whether |a-b|<e;

If |a-b|<e, then execute the described first test point and second test point to close all non-executive data processing devices, wherein, e is the second predetermined value; if |a-b|>=e, then judge whether i>=d;

If I>=d, then execute the described first test point and the second test point to close all non-executive data processing devices, wherein, I is the number of cycles, the initial value is 0, and d is the third predetermined value;

If I<d, and A>B, a<b, then calculate I=I+1, Y=a, a=b, b=X+Y-a; A=B, and then execute the b data The processing device executes all the threads in the multi-computer system, and measures the power consumption B of the multi-computer system;

If I<d, and A>B, a>b, then calculate I=I+1, X=a, a=b, b=X+Y-a; A=B, and then execute the b data The processing device executes all the threads in the multi-computer system, and measures the power consumption B of the multi-computer system;

If I<d, and A<B, a<b, then calculate I=I+1, X=b, a=a, b=X+Y-a; then execute the described b data processing equipment to execute the All threads in the multi-machine system, and measure the power consumption B of the multi-machine system;

If I<d, and A<B, a>b, then calculate I=I+1, Y=b, a=a, b=X+Y-a; then execute the b data processing equipment to execute the All the threads in the multi-computer system, and measure the power consumption B of the multi-computer system.

In the above step 13, closing all non-executive data processing devices according to the first test point and the second test point may include:

Calculate Z-(a+b)/2, where Z is the total number of data processing devices in the multi-computer system;

Shut down Z-(a+b)/2 data processing devices in the multi-computer system.

Taking a multi-core processor with N cores as an example, the method for optimizing power consumption of a multi-computer system will be further described in detail below.

When the number of cores of the test point is executed each time, if the number of cores of the test point is greater than the number of cores executed in the previous test point, it is necessary to migrate a part of the threads waiting for execution to the started and idle processor cores; if the number of cores of the test point If it is smaller than the number of cores executed in the previous test point, it is necessary to migrate the threads executed on the processor core that is about to be shut down to other active processor cores, and try to achieve an even load distribution of each active processor core.

Referring to Figure 2, the method for optimizing the power consumption of a multi-core processor includes the following steps:

Step 21. Fully parallelize the program running in the multi-core processor, and determine the initial number range [Y, X] of the multi-core processor for adjusting the cores of the multi-core processor, that is, determine the initial values of X and Y. Also, set I=0, where I is the number of cycles.

Among them, X is the upper bound of the number of cores, and Y is the lower bound of the number of cores.

For example, if the total number of threads of the running program is greater than the total number N of cores in the multi-core processor, the initial number of cores is determined to range from 0 to the total number of cores, that is, [0, N], that is, X=N, Y=0 .

If the total number of threads of the running program is less than the total number N of cores in the multi-core processor, the initial number of cores ranges from 0 to the total number of threads, that is, [0, the total number of threads], that is, X=the total number of threads, Y=0.

If the total number of threads is less than the total number N of cores in the multi-core processor, a method similar to traditional turbo boost can be used to turn off idle cores in the multi-core processor, and then perform steps 22 and later on the remaining cores to power the multi-core processor optimization.

Step 22. Calculate the first test point a=(X+Y)×m, and calculate the second test point b=(X+Y)×(1-m). Among them, the selection interval of m is (0, 1).

Here, (Y, X) is used as the search space for the optimal number of execution cores, and the first test point a and the second test point b are respectively selected in the search space as the number of test cores in this search.

Step 23, using a core in the multi-core processor as an execution core, migrating threads on other X-a cores in the multi-core processor to this a core, these X-a cores become non-executive cores, and shut down these non-executive core. Then, the frequency of each execution core is gradually reduced until the multi-core processor just meets the target performance, and the power consumption A of the multi-core processor at this time is measured. Assuming that all cores must run at the same frequency, when the frequency is reduced, the frequency of all cores is reduced together.

Wherein, the target performance is the target performance used after the power consumption management mechanism is started, and can be set in advance. For example: you can set a fixed target performance, you can set a fixed target performance lower limit, you can also set an upper limit of allowable performance loss, or you can use the directly measured performance of the multi-core processor when all available cores are normally executed as the target performance.

Step 24, using b cores in the multi-core processor as execution cores, migrating threads on other X-b cores in the multi-core processor to the b cores, these X-b cores become non-executive cores, and shut down these non-executive core. Then, the frequency of each execution core is gradually reduced until the multi-core processor just meets the target performance, and the power consumption B of the multiple processors at this time is measured.

Step 25. Determine whether the difference between power consumption A and power consumption B is less than a predetermined value, for example, determine whether |A-B|<w, if yes, go to step 28; otherwise, go to step 26. Among them, w is the predetermined value of the difference between A and B.

Step 26. Judging whether the difference between a and b is less than a predetermined value, such as judging whether |a-b|<ε, if yes, go to step 28; otherwise, go to step 27. Wherein, ε is a predetermined value of the difference.

Step 27. So far, whether the search process has reached a limit on the number of cycles, if it is judged whether I≥d, if yes, go to step 28; otherwise, go to step 29. Wherein, d is a predetermined numerical value of the number of cycles.

Step 28. Obtain that the number of cores required for the minimum power consumption of the multi-core processor is (a+b)/2, that is, determine the optimal number of execution cores in the multi-core processor to be (a+b)/2. Therefore, only ( a+b)/2 cores are used as execution cores, and all other non-execution cores are turned off, and the frequency of execution cores is gradually reduced until the target performance of the multi-core processor is just met, and the power consumption optimization of the multi-core processor is completed. The frequency at this point is the processor frequency required for the lowest power consumption. End the process.

Step 29, calculate I=I+1.

Step 210 , compare power consumption A and power consumption B, and determine whether power consumption A>power consumption B, if yes, execute step 211 ; otherwise, execute step 214 .

Step 211 , judging whether a<b, if yes, execute step 212 ; otherwise, execute step 213 .

Step 212: Reduce the new core number search interval to (a, X), that is, set the lower bound Y of the core number to a, that is to say, Y=a, and set the current b to be in the new core number search interval The first test point a is a=b, and b=X+Y-a. In this way, there is no need to perform the above step 23 afterwards, and A=B can be obtained directly, and then, step 24 is performed to obtain a new value of B.

Step 213: Reduce the new core number search interval to (Y, a), that is, set the upper bound X of the core number to a, that is to say, X=a, and set b to be the first in the new core number search interval The test core number a is a=b. And let b=X+Y-a. In this way, there is no need to perform the above step 23 afterwards, and A=B can be obtained directly, and then, step 24 is performed to obtain a new value of B.

Step 214 , judging whether a<b, if yes, execute step 215 ; otherwise, execute step 216 .

Step 215. Reduce the new core number search interval to (Y, b), that is, set the upper bound X of the core number to b, that is to say, X=b, and set a to be the first in the new core number search interval The number of test cores a is a=a. And let b=X+Y-a. In this way, the value of A remains unchanged, and the above step 23 does not need to be performed afterwards. Then, execute step 24 to obtain a new B value.

Step 216: Reduce the new core number search interval to (b, X), that is, set the lower bound Y of the core number to b, that is to say, Y=b, and set a to be the first in the new core number search interval The test core number a is a=a. And let b=X+Y-a. In this way, the value of A remains unchanged, and the above step 23 does not need to be performed afterwards. Then, execute step 24 to obtain a new B value.

After the above step 28 is executed and the power consumption of the multi-machine system reaches the minimum, if any of the following situations exists, the above steps 21 to 216 can be performed again to optimize the power consumption of the multi-machine system:

a. The number of threads run by the multi-core processor changes;

b. The load of the multi-threaded program changes;

c. When the power consumption management mechanism starts;

d. When changing the set target performance.

In the above embodiment, by determining two test points a and b for each search, the interval on the side of the test point with a larger power consumption value will be discarded after each search, and the test point will be used as the next search core. The boundary of the number range, thus effectively narrowing the search range of the number of cores. And, the search range of the number of cores is continuously narrowed by performing the search in a loop until the condition for stopping the search is reached. In this way, if the fraction m used in each search in the loop is the same, the test point that is not used as a boundary in the next search can still be used as a test point, and the power consumption obtained from the test can directly use the result measured last time. Save performing measurements once per loop.

Using the technical solutions provided by the above embodiments, firstly, the search interval that needs to be discarded can be quickly judged in the space of power consumption-number of cores, and a power consumption value measured last time can be used for power consumption comparison of each test point, Reduces the overhead of searching for measurements and quickly reduces the space that needs to be searched. Secondly, while finding the core number and frequency required for the lowest power consumption, you can turn off the processor cores that do not need to execute the program to save static power consumption, or execute other programs on the processor cores that do not need to execute the program , greatly improving the utilization rate of the processor and improving the energy efficiency of the processor. Finally, a special processor can be designed for a certain type of application, and the number of cores and frequency required for the lowest power consumption can be found according to the characteristics of the program to guide the design of the chip, which can effectively control the chip size and greatly save on-chip hardware costs.

FIG. 3 is a schematic structural diagram of an apparatus for optimizing power consumption of a multi-computer system provided by an embodiment of the present invention. The device provided in this embodiment is used to implement the method shown in FIG. 1 above. As shown in FIG. 3 , the device includes: a range determination unit 31 , a test point determination unit 32 and a performance adjustment unit 33 .

The range determination unit 31 is used to determine the range of the number of data processing devices used to adjust the power consumption of the multi-computer system in the multi-computer system, the minimum value of the number range is the lower limit, and the maximum value is the upper limit.

The test point determination unit 32 is used to search within the quantity range to determine the first test point and the second test point; the first test point and the second test point are the quantity of data processing equipment, and the first test point The sum of the first test point and the second test point is equal to the sum of the upper bound and the lower bound.

The performance adjustment unit 33 is configured to shut down all non-executing data processing devices according to the first test point and the second test point, and gradually reduce the frequency of the remaining data processing devices to meet the target performance.

The range determining unit 31 may be specifically configured to determine an initial quantity range of data processing devices used to adjust power consumption of the multi-computer system in the multi-computer system.

Further, the range determination unit 31 can be specifically used for:

If the number of threads running in the multi-computer system is greater than the number of all data processing devices in the multi-computer system, the minimum value of the initial number range is 0, and the maximum value is all data processing devices in the multi-computer system quantity;

If the total number of threads or the number of running threads in the multi-computer system is less than the number of all data processing devices in the multi-computer system, the minimum value of the initial number range is 0, and the maximum value is 0 in the multi-computer system The number of threads running in .

Optionally, the test point determination unit 32 can be specifically used for:

Calculate a=(X+Y)×m, b=(X+Y)×(1-m); where a is the first test point, b is the second test point, 0<m<1 , X is the upper bound, Y is the lower bound, and a, b, X, and Y are variables;

Execute all threads in the multi-computer system with the a data processing device, and measure the power consumption A of the multi-computer system;

Execute all threads in the multi-computer system with the b data processing devices, and measure the power consumption B of the multi-computer system;

compare said A and B;

If |A-B|<w, execute the step of shutting down all non-executing data processing devices according to the first test point and the second test point, where w is a first predetermined value;

If |A-B|>=w, then judge whether |a-b|<e;

If |a-b|<e, then execute the described first test point and second test point to close all non-executive data processing devices, wherein, e is the second predetermined value; if |a-b|>=e, then judge whether i>=d;

If I>=d, then execute the described first test point and the second test point to close all non-executive data processing devices, wherein, I is the number of cycles, the initial value is 0, and d is the third predetermined value;

If I<d, and A>B, a<b, then calculate I=I+1, Y=a, a=b, b=X+Y-a; A=B, and then execute the b data The processing device executes all the threads in the multi-computer system, and measures the power consumption B of the multi-computer system;

If I<d, and A>B, a>b, then calculate I=I+1, X=a, a=b, b=X+Y-a; A=B, and then execute the b data The processing device executes all the threads in the multi-computer system, and measures the power consumption B of the multi-computer system;

If I<d, and A<B, a<b, then calculate I=I+1, X=b, a=a, b=X+Y-a; then execute the described b data processing equipment to execute the All threads in the multi-machine system, and measure the power consumption B of the multi-machine system;

If I<d, and A<B, a>b, then calculate I=I+1, Y=b, a=a, b=X+Y-a; then execute the b data processing equipment to execute the All the threads in the multi-computer system, and measure the power consumption B of the multi-computer system.

Optionally, the performance adjustment unit 33 may include: a quantity calculation subunit 331 and a device shutdown subunit 332 .

The quantity calculation subunit 331 is used to calculate Z-(a+b)/2, where Z is the total number of data processing devices in the multi-computer system;

The device shutting down subunit 332 is used to shut down Z-(a+b)/2 data processing devices in the multi-computer system.

FIG. 4 is a schematic structural diagram of a multi-machine system according to an embodiment of the present invention. As shown in FIG. 4 , the multi-machine system can add a power consumption optimization function on the basis of existing multi-core processors, multi-processor systems, multi-device systems, etc., that is, it includes a multi-machine system body 41 and an optimization device 42 . The multi-machine system body 41 may be an existing multi-core processor, a multi-processor system, a multi-device system, and the like. The optimization device 42 may be any device shown in FIG. 3 for optimizing the power consumption of the multi-machine system, and is used for optimizing the power consumption of the multi-machine system body 41 .

The embodiment of the present invention also provides a computer program product, the computer program product includes a computer-readable medium, and the readable medium includes a first set of program codes for executing the steps in the method shown in Figure 1 above:

Determining the number range of data processing devices used to adjust the power consumption of the multi-machine system in the multi-machine system, the minimum value of the number range is the lower bound, and the maximum value is the upper bound;

Search within the number range to determine the first test point and the second test point; the first test point and the second test point are the number of data processing equipment, and the first test point and the second test point The sum of the two test points is equal to the sum of the upper bound and the lower bound;

Shut down all non-executing data processing devices according to the first test point and the second test point, and gradually reduce the frequency of the remaining data processing devices to meet the target performance.

Optionally, determining the number range of data processing devices used to adjust the power consumption of the multi-computer system in the multi-computer system includes:

Determine an initial quantity range of data processing devices in the multi-computer system for adjusting power consumption of the multi-computer system.

Optionally, determining an initial quantity range of data processing devices used to adjust power consumption of the multi-computer system in the multi-computer system includes:

If the number of threads running in the multi-computer system is greater than the number of all data processing devices in the multi-computer system, the minimum value of the initial number range is 0, and the maximum value is all data processing devices in the multi-computer system quantity;

If the total number of threads or the number of running threads in the multi-computer system is less than the number of all data processing devices in the multi-computer system, the minimum value of the initial number range is 0, and the maximum value is 0 in the multi-computer system The number of threads running in .

Optionally, searching within the quantity range to determine the first test point and the second test point includes:

Calculate a=(X+Y)×m, b=(X+Y)×(1-m); where a is the first test point, b is the second test point, 0<m<1 , X is the upper bound, Y is the lower bound, and a, b, X, and Y are variables;

Execute all threads in the multi-computer system with the a data processing device, and measure the power consumption A of the multi-computer system;

Execute all threads in the multi-computer system with the b data processing devices, and measure the power consumption B of the multi-computer system;

compare said A and B;

If |A-B|<w, execute the step of shutting down all non-executing data processing devices according to the first test point and the second test point, where w is a first predetermined value;

If |A-B|>=w, judge whether |a-b|<e;

If |a-b|<e, then perform the described first test point and second test point to close all non-executive data processing devices, wherein, e is the second predetermined value; if |a-b|>=e, then judge whether i>=d;

If I>=d, then execute the described first test point and the second test point to close all non-executive data processing devices, wherein, I is the number of cycles, the initial value is 0, and d is the third predetermined value;

If I<d, and A>B, a<b, then calculate I=I+1, Y=a, a=b, b=X+Y-a; A=B, and then execute the b data The processing device executes all the threads in the multi-computer system, and measures the power consumption B of the multi-computer system;

If I<d, and A>B, a>b, then calculate I=I+1, X=a, a=b, b=X+Y-a; A=B, and then execute the b data The processing device executes all the threads in the multi-computer system, and measures the power consumption B of the multi-computer system;

If I<d, and A<B, a<b, then calculate I=I+1, X=b, a=a, b=X+Y-a; then execute the b data processing equipment to execute the All threads in the multi-machine system, and measure the power consumption B of the multi-machine system;

If I<d, and A<B, a>b, then calculate I=I+1, Y=b, a=a, b=X+Y-a; then execute the b data processing equipment to execute the All the threads in the multi-computer system, and measure the power consumption B of the multi-computer system.

Optionally, shutting down all non-executive data processing devices according to the first test point and the second test point includes:

Calculate Z-(a+b)/2, where Z is the total number of data processing devices in the multi-computer system;

Shut down Z-(a+b)/2 data processing devices in the multi-computer system.

Those of ordinary skill in the art can understand that all or part of the steps for implementing the above method embodiments can be completed by program instructions and related hardware. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps including the above-mentioned method embodiments; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk and other various media that can store program codes.

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than limiting them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: It is still possible to modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the technical solutions of the various embodiments of the present invention. scope.

Claims (11)

1. A method for optimizing power consumption of a multi-machine system, comprising: Determining the number range of data processing devices used to adjust the power consumption of the multi-machine system in the multi-machine system, the minimum value of the number range is the lower bound, and the maximum value is the upper bound; Search within the number range to determine the first test point and the second test point; the first test point and the second test point are the number of data processing equipment, and the first test point and the second test point The sum of the two test points is equal to the sum of the upper bound and the lower bound; Shutting down all non-executing data processing devices according to said first test point and second test point, and gradually reducing the frequency of remaining data processing devices to meet target performance; Wherein, said searching within said quantity range to determine the first test point and the second test point includes: Calculate a=(X+Y)×m, b=(X+Y)×(1-m); wherein, a is the first test point, b is the second test point, 0<m<1 , X is the upper bound, Y is the lower bound, and a, b, X, and Y are variables; Execute all the threads in the multi-computer system with a data processing device, and measure the power consumption A of the multi-computer system; Execute all threads in the multi-computer system with b data processing devices, and measure the power consumption B of the multi-computer system; compare said A and B; If |A-B|<w, execute the step of shutting down all non-executing data processing devices according to the first test point and the second test point, where w is a first predetermined value; If |A-B|>=w, judge whether |a-b|<e; If |a-b|<e, execute the step of shutting down all non-executing data processing devices according to the first test point and the second test point, where e is a second predetermined value; If |a-b|>=e, then judge whether I>=d, wherein, I is the number of cycles, the initial value is 0, and d is the third predetermined value; If I>=d, then execute the described first test point and the second test point to close all non-executive data processing devices; if I<d, then according to the size relationship between A and B, and the size of a and b relationship, to narrow down the range of quantities stated. 2. The method according to claim 1, wherein determining the number range of the data processing equipment used to adjust the power consumption of the multi-machine system in the multi-machine system comprises: Determine an initial quantity range of data processing devices in the multi-computer system for adjusting power consumption of the multi-computer system. 3. The method according to claim 2, wherein determining the initial quantity range of the data processing equipment used to adjust the power consumption of the multi-machine system in the multi-machine system comprises: If the number of threads running in the multi-computer system is greater than the number of all data processing devices in the multi-computer system, the minimum value of the initial number range is 0, and the maximum value is all data processing devices in the multi-computer system quantity; If the total number of threads or the number of running threads in the multi-computer system is less than the number of all data processing devices in the multi-computer system, the minimum value of the initial number range is 0, and the maximum value is 0 in the multi-computer system The number of threads running in . 4. according to the described method of any one of claim 1-3, it is characterized in that, described if I<d, then according to the magnitude relation of A and B, and the magnitude relation of a and b, narrow down described quantity scope, comprise : If I<d, and A>B, a<b, then calculate I=I+1, Y=a, a=b, b=X+Y-a; A=B, then carry out described with b data processing equipment Execute all threads in the multi-computer system, and measure the power consumption B of the multi-computer system; If I<d, and A>B, a>b, then calculate I=I+1, X=a, a=b, b=X+Y-a; A=B, then carry out described with b data processing equipment Execute all threads in the multi-computer system, and measure the power consumption B of the multi-computer system; If I<d, and A<B, a<b, then calculate I=I+1, X=b, a=a, b=X+Y-a; then execute the described multiprocessing with b data processing devices All the threads in the machine system, and measure the power consumption B of the multi-machine system; If I<d, and A<B, a>b, then calculate I=I+1, Y=b, a=a, b=X+Y-a; then execute the b data processing equipment to execute the multiple All the threads in the multi-machine system, and measure the power consumption B of the multi-machine system. 5. The method according to claim 4, characterized in that, closing all non-executive data processing devices according to the first test point and the second test point, comprising: Calculate Z-(a+b)/2, where Z is the total number of data processing devices in the multi-computer system; Shut down Z-(a+b)/2 data processing devices in the multi-computer system. 6. A device for optimizing power consumption of a multi-computer system, comprising: A range determination unit, configured to determine the range of the number of data processing devices used to adjust the power consumption of the multi-machine system in the multi-computer system, the minimum value of the number range is a lower bound, and the maximum value is an upper bound; The test point determination unit is used to search within the quantity range to determine the first test point and the second test point; the first test point and the second test point are the numbers of data processing equipment, and the first test point The sum of a test point and said second test point is equal to the sum of said upper bound and said lower bound; A performance adjustment unit, configured to shut down all non-executing data processing devices according to the first test point and the second test point, and gradually reduce the frequency of the remaining data processing devices to meet the target performance; The test point determination unit is specifically used for: Calculate a=(X+Y)×m, b=(X+Y)×(1-m); wherein, a is the first test point, b is the second test point, 0<m<1 , X is the upper bound, Y is the lower bound, and a, b, X, and Y are variables; Execute all the threads in the multi-computer system with a data processing device, and measure the power consumption A of the multi-computer system; Execute all threads in the multi-computer system with b data processing devices, and measure the power consumption B of the multi-computer system; compare said A and B; If |A-B|<w, execute the step of shutting down all non-executing data processing devices according to the first test point and the second test point, where w is a first predetermined value; If |A-B|>=w, judge whether |a-b|<e; If |a-b|<e, execute the described first test point and second test point to close all non-executive data processing devices, wherein, e is the second predetermined value; if |a-b|>=e, then judge Whether i>=d; If I>=d, then carry out described closing all non-executive data processing devices according to the first test point and the second test point, wherein, I is the number of cycles, the initial value is 0, and d is the third predetermined value; if If I<d, the quantity range is reduced according to the size relationship between A and B, and the size relationship between a and b. 7 . The device according to claim 6 , wherein the range determination unit is specifically configured to determine an initial quantity range of data processing devices in the multi-computer system for adjusting power consumption of the multi-computer system. 8 . 8. The device according to claim 7, wherein the range determining unit is specifically used for: If the number of threads running in the multi-computer system is greater than the number of all data processing devices in the multi-computer system, the minimum value of the initial number range is 0, and the maximum value is all data processing devices in the multi-computer system quantity; If the total number of threads or the number of running threads in the multi-computer system is less than the number of all data processing devices in the multi-computer system, the minimum value of the initial number range is 0, and the maximum value is 0 in the multi-computer system The number of threads running in . 9. The device according to any one of claims 6-8, wherein the test point determination unit is used to reduce the size of the test point according to the size relationship between A and B and the size relationship between a and b if I<d The stated quantity range includes: If I<d, and A>B, a<b, then calculate I=I+1, Y=a, a=b, b=X+Y-a; A=B, then carry out described with b data processing equipment Execute all threads in the multi-machine system; If I<d, and A>B, a>b, then calculate I=I+1, X=a, a=b, b=X+Y-a; A=B, then carry out described with b data processing equipment Execute all threads in the multi-machine system; If I<d, and A<B, a<b, then calculate I=I+1, X=b, a=a, b=X+Y-a; then execute the described multiprocessing with b data processing devices all threads in the machine system; If I<d, and A<B, a>b, then calculate I=I+1, Y=b, a=a, b=X+Y-a; then execute the b data processing equipment to execute the multiple all threads in the computer system. 10. The device according to claim 9, wherein the performance adjustment unit comprises: A quantity calculation subunit, used to calculate Z-(a+b)/2, where Z is the total number of data processing devices in the multi-computer system; The device shutting down subunit is used to shut down Z-(a+b)/2 data processing devices in the multi-computer system. 11. A multi-machine system, characterized by comprising a multi-machine system body and the device for optimizing the power consumption of the multi-machine system according to any one of claims 6-10, the device is used for optimizing the multi-machine system The power consumption of the machine system body.