GPU cluster starting system based on use case

Abstract

The invention relates to the technical field of GPU cluster optimization, in particular to a GPU cluster starting system based on a use case, the system comprises a GPU cluster, a processor and a memory in which a computer program is stored, and when the computer program is executed by the processor, the following steps are realized: when a request for executing a target use case is received, the GPU cluster is used for executing a preset use case, when the first preset condition is met, the GPU cluster is used for executing the target use case and completing starting of the GPU cluster, the GPU cluster firstly executes the preset use case to control the hardware resource occupancy rate of the GPU cluster and then starts and executes the target use case, the target use case can run at the maximum power after starting and execution, it is guaranteed that the performance loss is small, and the efficiency is high. And the instantaneous current change during starting execution of the target use case and the preset use case is not too large, so that the stability and the safety during starting of the GPU cluster are improved.

Description

Technical Field

[0001] This invention relates to the field of GPU cluster optimization technology, and in particular to a use case-based GPU cluster startup system. Background Technology

[0002] In the use case of large-scale GPU clusters, when a single GPU chip changes from an idle state to a running state, the power consumption of the GPU chip will instantly rise from a low value to a high value, for example, from 50W to 300W. Since a large-scale GPU cluster contains a large number of GPU chips, such as in the case of a multi-card cluster, when the large-scale GPU cluster changes from an idle state to a running state, the power consumption of all the GPU chips contained in it needs to increase instantaneously, resulting in a significant increase in the demand for instantaneous current.

[0003] The above situations may cause the instantaneous current demand to exceed the power supply system's carrying capacity, leading to power overload, which in turn may cause power outages or voltage drops, resulting in unstable power supply or even power outages. Moreover, when the instantaneous current exceeds the threshold limit set by the circuit design, it will trigger the circuit breaker or fuse to automatically cut off the current to prevent the circuit from overheating or the equipment from being damaged, which may cause the GPU chip to fail to start and run normally. In addition, the simultaneous startup of a large number of GPU chips will cause the voltage of the power supply line to drop sharply, which may affect the performance and stability of other devices on the power supply line. Furthermore, current overload may also cause hardware components to overheat, such as the motherboard, power supply unit, and GPU chip, accelerating hardware aging or even damage. Finally, the simultaneous startup of a large number of GPU chips also means that there is a greater demand for heat dissipation, which increases the pressure on the cooling system.

[0004] One way to solve the above problem is to gradually increase the operating frequency and voltage by controlling the DPM level, thereby gradually increasing the current of the GPU cluster and ensuring that the instantaneous current during DPM level switching is not too large. However, although this method can effectively reduce the instantaneous current surge when the GPU cluster starts up, there will be performance loss when executing test cases. Therefore, how to reduce the instantaneous current surge when the GPU cluster starts up while ensuring minimal performance loss, and thus improve the stability and security of the GPU cluster during startup, has become an urgent problem to be solved. Summary of the Invention

[0005] To address the aforementioned technical problems, the technical solution adopted by this invention is as follows:

[0006] A use case-based GPU cluster startup system includes: a GPU cluster, a processor, and a memory storing a computer program. The memory further includes a preset use case K, which corresponds to a preset occupancy rate Q. When the computer program is executed by the processor, the following steps are implemented:

[0007] S101, when a request to execute a target use case using the GPU cluster is received, the preset use case K is executed using the GPU cluster. When the preset use case K is executed using the GPU cluster, the first hardware resource utilization rate of the GPU cluster is the preset utilization rate Q.

[0008] S102, when the first preset condition is met, the GPU cluster is used to execute the target use case and the GPU cluster is started. When the GPU cluster is used to execute the target use case, the second hardware resource utilization rate of the GPU cluster is P, where P and Q are both values ​​in the range of [0,1], and P is greater than Q.

[0009] Compared with the prior art, the present invention has significant advantages. Through the above technical solution, the GPU cluster startup system based on use cases provided by the present invention achieves considerable technical progress and practicality, and has broad industrial application value. It has at least the following advantages:

[0010] This invention provides a use case-based GPU cluster startup system. The system includes a GPU cluster, a processor, and a memory storing a computer program. The memory further includes a preset use case K, which corresponds to a preset utilization rate Q. When the computer program is executed by the processor, the following steps are implemented: when a request to execute a target use case using the GPU cluster is received, the preset use case K is executed using the GPU cluster. When the preset use case K is executed using the GPU cluster, the first hardware resource utilization rate of the GPU cluster is the preset utilization rate Q. When a first preset condition is met, the target use case is executed using the GPU cluster to complete the startup of the GPU cluster. When the target use case is executed using the GPU cluster, the second hardware resource utilization rate of the GPU cluster is P. Both P and Q are values ​​in the range [0,1], and P is greater than Q.

[0011] It is known that before the GPU cluster starts executing the target test case, the GPU cluster first executes a preset test case to control the hardware resource utilization rate of the GPU cluster to reach the preset utilization rate, thereby keeping the instantaneous current surge during the execution of the preset test case within an acceptable range. Then, the target test case is started and executed. After the target test case starts, it can run at maximum power, ensuring that the performance loss of the GPU cluster during the execution of the target test case is small. In addition, the current of the GPU cluster during startup can be gradually increased, ensuring that the instantaneous current changes during the execution of the target test case and the preset test case are not too large, reducing the instantaneous current surge during the startup of the GPU cluster, thereby improving the stability and security of the GPU cluster during startup. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart illustrating the process of a computer program being executed by a processor in a GPU cluster startup system based on a use case, as provided in an embodiment of the present invention. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] This embodiment provides a use case-based GPU cluster startup system. The system includes a GPU cluster, a processor, and a memory storing computer programs. The memory also includes a preset use case K, which corresponds to a preset occupancy rate Q. (See [link to documentation]). Figure 1 This is a flowchart illustrating the execution of a computer program by a processor in a GPU cluster startup system based on a use case, according to an embodiment of the present invention. When the computer program is executed by the processor, the following steps are implemented:

[0016] S101, when a request to execute a target use case using the GPU cluster is received, the preset use case K is executed using the GPU cluster. When the preset use case K is executed using the GPU cluster, the first hardware resource utilization rate of the GPU cluster is the preset utilization rate Q.

[0017] S102, when the first preset condition is met, the GPU cluster is used to execute the target use case and the GPU cluster is started. When the GPU cluster is used to execute the target use case, the second hardware resource utilization rate of the GPU cluster is P, where P and Q are both values ​​in the range of [0,1], and P is greater than Q.

[0018] The target use case can be provided by the user side. When the target use case is executed using the GPU cluster, the second hardware resource utilization rate of the GPU cluster is P. The preset use case can be provided by the device side. When the target use case is executed using the GPU cluster, the second hardware resource utilization rate of the GPU cluster is Q. Since the preset use case can be written in advance or provided by the user, the preset utilization rate Q corresponding to the preset use case can be determined. It can be seen that the startup instantaneous current surge corresponding to the execution of the preset use case with the preset utilization rate Q is within an acceptable range. The hardware resource utilization rate can refer to the utilization rate of all chips in the GPU cluster. For example, the preset utilization rate Q can be 0.5. At this time, only half of the service resources of each GPU chip in the GPU cluster are used. In this embodiment, the size of the instantaneous current surge is controlled by controlling the utilization rate of all chips in the GPU cluster.

[0019] Specifically, when the second hardware resource utilization rate P is less than or equal to Q when using the GPU cluster to execute the target use case, the GPU cluster can be used directly to execute the target use case. Otherwise, the GPU cluster needs to be used to execute the preset use case K. When the first preset condition is met, the GPU cluster is used to execute the target use case. In this method, the preset use case is equivalent to a warm-up effect for the GPU cluster. When the GPU cluster changes from an idle state to the state of executing the preset use case, the instantaneous current surge is within an acceptable range. When the GPU cluster changes from the state of executing the preset use case to the state of executing the target use case, the instantaneous current surge is determined based on the operating current of the GPU cluster when executing the preset use case and the operating current when executing the target use case, rather than the current of the GPU cluster in the idle state and the operating current when executing the target use case. This achieves the effect of gradually increasing the current, ensuring that the instantaneous current changes when the target use case and the preset use case are started and executed are not too large, reducing the instantaneous current surge when the GPU cluster starts, and thus improving the stability and security of the GPU cluster during startup.

[0020] In one specific implementation, the first preset condition is: the execution time of the preset use case using the GPU cluster is greater than a preset first duration threshold or the preset use case has been completed.

[0021] If the execution time of the preset test case using the GPU cluster is zero, it is equivalent to directly using the GPU cluster to start and execute the target test case. In other words, the execution time of the preset test case using the GPU cluster is zero, which is the traditional approach. Therefore, in this embodiment, the first time threshold should be a number greater than zero.

[0022] Specifically, introducing a first duration threshold or the time taken for the preset test case to complete execution between the execution of the preset test case and the target test case facilitates adaptation to the response time of the power system and ensures the stability of the power system supplying power to the GPU cluster.

[0023] In one specific implementation, the difference between the current time and the start time of executing the preset use case using the GPU cluster is used as the execution duration.

[0024] In one embodiment, the first duration threshold t1 is obtained by mapping the power system redundancy C through a preset mapping function.

[0025] The power system redundancy C can be obtained from the power system side. In this embodiment, for ease of representation, the power system redundancy C is represented by a normalized value. When the power system redundancy C is 0, it means that the power system has no redundancy, and the power system response time is the longest. When the power system redundancy C is 1, it means that the power system is completely redundant, and the power system response time is the shortest. That is, the larger the power system redundancy, the shorter the power system response time, and the smaller the power system redundancy, the longer the power system response time.

[0026] Specifically, the preset mapping function is: t1=(F / (e rC-b +1))×T, where F=e -b +1, r is the first adjustment coefficient, b is the second adjustment coefficient, and T is the theoretical maximum delay. Here, r can be a value greater than one. In this embodiment, r is set to 5, b is set to 0.5, k and b can be used to adjust the slope of the mapping function. Implementers can adjust the values ​​of r and b according to actual needs. T is the theoretical maximum delay. When C = 0, t = T; when C = 1, t approaches 0. Since the mapping result of this mapping function is always greater than zero, it also ensures that the first duration threshold should be a number greater than zero.

[0027] In one specific implementation, the preset use case K includes M preset sub-use cases {k1, k2, ..., k}. m , ..., k M}, where k m Let m be the m-th preset sub-use case in the preset use case K, where m is an integer in the range [1, M].

[0028] In one specific implementation, k m Corresponding to the reference occupancy rate q m q m For values ​​in the range [0, 1], the step of using the GPU cluster to execute the preset use case K includes:

[0029] S201, Initialize test case identifier n=1;

[0030] S202, with k n As a current use case, k nThis is the nth preset sub-use case in the preset use case K;

[0031] S203, execute the current use case using the GPU cluster. When executing the current use case using the GPU cluster, the third hardware resource utilization rate corresponding to the GPU cluster is q. n q n For k n Corresponding reference occupancy rate;

[0032] S204, when the second preset condition is met, update n = n+1, return to execute steps S202 to S204, until n = M+1, and execute step S102.

[0033] In one specific implementation, the second preset condition is: the execution time of the current use case using the GPU cluster is greater than a preset second duration threshold, or the current use case has been completed.

[0034] The method for determining the second duration threshold can be the same as the method for determining the first duration threshold. The implementer only needs to adjust the values ​​of T, r, and b in the preset mapping function, which will not be elaborated here.

[0035] In one specific implementation, q i i+1 , where q i For the i-th preset sub-use case k in the preset use case K i The corresponding reference occupancy rate, q i+1 For the (i+1)th preset sub-use case k in the preset use case K i+1 The corresponding reference occupancy rate, where i is an integer in the range [1, M-1].

[0036] In order to ensure that the current of the GPU cluster gradually increases when switching between preset sub-use cases, constraint q is applied. i i+1 In other words, the reference occupancy rate of the earlier preset sub-use cases is less than the reference occupancy rate of the later preset sub-use cases.

[0037] In one specific implementation, the characteristic is that M = 10.

[0038] In this embodiment, the number of preset sub-use cases in the preset use cases can be set to 10. The implementer can adjust the number of preset sub-use cases in the preset use cases according to the actual situation.

[0039] ​​In this embodiment, before the GPU cluster starts executing the target use case, the GPU cluster first executes a preset use case to control the hardware resource utilization rate of the GPU cluster to reach a preset utilization rate. This ensures that the instantaneous current surge when starting the preset use case is within an acceptable range. Then, the target use case is started and executed. After the target use case starts, it can run at maximum power, ensuring that the performance loss of the GPU cluster when executing the target use case is small. In addition, the current of the GPU cluster at startup can be gradually increased, ensuring that the instantaneous current changes when the target use case and the preset use case are started are not too large, reducing the instantaneous current surge when the GPU cluster starts, thereby improving the stability and security of the GPU cluster at startup.

[0040] While specific embodiments of the invention have been described in detail by way of example, those skilled in the art should understand that the above examples are for illustrative purposes only and are not intended to limit the scope of the invention. Those skilled in the art should also understand that various modifications can be made to the embodiments without departing from the scope and spirit of the invention. The scope of the invention is defined by the appended claims.

Claims

1. A use case-based GPU cluster startup system, characterized in that, The system includes a GPU cluster, a processor, and a memory storing a computer program. The memory also includes a preset use case K, which corresponds to a preset occupancy rate Q. When the computer program is executed by the processor, the following steps are performed: S101, when a request to execute a target use case using the GPU cluster is received, the preset use case K is executed using the GPU cluster. When the preset use case K is executed using the GPU cluster, the first hardware resource utilization rate of the GPU cluster is the preset utilization rate Q. S102, when the first preset condition is met, the GPU cluster is used to execute the target use case and the GPU cluster is started. When the GPU cluster is used to execute the target use case, the second hardware resource utilization rate of the GPU cluster is P, where P and Q are both values ​​in the range of [0,1], and P is greater than Q.

2. The use case-based GPU cluster startup system according to claim 1, characterized in that, The first preset condition is: the execution time of the preset test case using the GPU cluster is greater than the preset first duration threshold, or the preset test case has been completed.

3. The use case-based GPU cluster startup system according to claim 2, characterized in that, The execution duration is determined by subtracting the start time of the preset use case from the current time when the GPU cluster is used to execute it.

4. The use case-based GPU cluster startup system according to claim 1, characterized in that, The preset use case K contains M preset sub-use cases {k1, k2, ..., k} m , ..., k M }, where k m Let m be the m-th preset sub-use case in the preset use case K, where m is an integer in the range [1, M].

5. The use case-based GPU cluster startup system according to claim 4, characterized in that, k m Corresponding to the reference occupancy rate q m q m For values ​​in the range [0, 1], the step of using the GPU cluster to execute the preset use case K includes: S201, Initialize test case identifier n=1; S202, with k n As a current use case, k n This is the nth preset sub-use case in the preset use case K; S203, execute the current use case using the GPU cluster. When executing the current use case using the GPU cluster, the third hardware resource utilization rate corresponding to the GPU cluster is q. n q n For k n Corresponding reference occupancy rate; S204, when the second preset condition is met, update n = n+1, return to execute steps S202 to S204, until n = M+1, and execute step S102.

6. The use case-based GPU cluster startup system according to claim 5, characterized in that, The second preset condition is: the execution time of the current use case using the GPU cluster is greater than the preset second duration threshold, or the current use case has been completed.

7. The use case-based GPU cluster startup system according to claim 5, characterized in that, q i i+1 , where q i For the i-th preset sub-use case k in the preset use case K i The corresponding reference occupancy rate, q i+1 For the (i+1)th preset sub-use case k in the preset use case K i+1 The corresponding reference occupancy rate, where i is an integer in the range [1, M-1].​ 8. The use case-based GPU cluster startup system according to claim 4, characterized in that, M＝10。

Images