Computing resource scheduling method and related device

Abstract

The embodiment of the invention discloses a computing resource scheduling method and a related device thereof. The temperature of each computing resource in a computing system is obtained; when it is detected that a target computing resource with the temperature exceeding a preset temperature threshold value exists in the computing system, a target computing task borne by the target computing resource is determined; determining a target scheduling strategy according to the task type of the target computing task and the resource type of the target computing resource; the task type is a complete machine task or a non-complete machine task, and the resource type is a computing device or a computing card; and finally, adjusting the working state of the target computing resource by adopting the target scheduling strategy. In this way, the temperature of the target computing resource is reduced, and serious negative effects caused by too high temperature are avoided.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a computing resource scheduling method and related devices. Background Art

[0002] With the continuous development of artificial intelligence (AI) technology, AI models are widely used in many businesses; and as the business requirements for related services increase, AI large models with large parameters and excellent performance are gradually emerging. Since AI large models have a large number of training parameters and a high scale of data exchange during training, it is necessary to use computing devices with large computing power to perform such training tasks, and usually rely on heterogeneous computing resources to perform such training tasks.

[0003] In actual applications, as the training task proceeds, the utilization rate of the computing card (such as the Graphics Processing Unit (GPU)) in the computing device continues to increase, and the temperature of the computing card will continue to rise, and the temperature of the computing card has a great impact on the model training task. For example, if the temperature of the computing card exceeds the temperature critical line, it will cause the computing device to which it belongs to restart, and the restart of the computing device will cause the model training task it carries to be terminated.

[0004] It can be seen that how to avoid the negative impact of temperature on computing cards in computing devices and ensure the stable operation of computing devices is a problem that needs to be solved urgently. Summary of the invention

[0005] The embodiments of the present application provide a computing resource scheduling method and related devices, which can use corresponding scheduling strategies to adjust the working state of the computing resource when the temperature of the computing resource exceeds the temperature threshold, thereby reducing the temperature of the computing resource and avoiding negative effects caused by excessive temperature.

[0006] The first aspect of the present application provides a computing resource scheduling method, the method comprising:

[0007] Acquire the temperature of each computing resource in the computing system; the computing resources include computing devices in the computing system and computing cards in the computing devices;

[0008] When detecting that there is a target computing resource in the computing system whose temperature exceeds a preset temperature threshold, determining a target computing task carried by the target computing resource;

[0009] Determine a target scheduling strategy according to the task type of the target computing task and the resource type of the target computing resource; the task type is a whole machine task or a non-whole machine task, and the resource type is a computing device or a computing card;

[0010] The target scheduling strategy is adopted to adjust the working state of the target computing resource.

[0011] A second aspect of the present application provides a computing resource scheduling device, the device comprising:

[0012] A temperature acquisition module, used to acquire the temperature of each computing resource in the computing system; the computing resources include computing devices in the computing system and computing cards in the computing devices;

[0013] A task determination module, configured to determine a target computing task carried by a target computing resource when it is detected that there is a target computing resource in the computing system whose temperature exceeds a preset temperature threshold;

[0014] A strategy determination module, used to determine a target scheduling strategy according to a task type of the target computing task and a resource type of the target computing resource; the task type is a whole machine task or a non-whole machine task, and the resource type is a computing device or a computing card;

[0015] The scheduling module is used to adopt the target scheduling strategy to adjust the working state of the target computing resource.

[0016] A third aspect of the present application provides a computer device, the device comprising a processor and a memory:

[0017] The memory is used to store computer programs;

[0018] The processor is used to execute the steps of the computing resource scheduling method as described in the first aspect according to the computer program.

[0019] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store a computer program, and the computer program is used to execute the steps of the computing resource scheduling method described in the first aspect above.

[0020] In a fifth aspect, the present application provides a computer program product or a computer program, the computer program product or the computer program includes a computer instruction, the computer instruction is stored in a computer-readable storage medium. A processor of a computer device reads the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device performs the steps of the computing resource scheduling method described in the first aspect above.

[0021] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0022] The embodiment of the present application provides a computing resource scheduling method, in which the temperature of each computing resource in the computing system is first obtained; wherein the computing resources include computing devices in the computing system and computing cards in the computing devices; when it is detected that there is a target computing resource in the computing system whose temperature exceeds a preset temperature threshold, the target computing task carried by the target computing resource is determined; then the target scheduling strategy is determined according to the task type of the target computing task and the resource type of the target computing resource; wherein the task type is a whole machine task or a non-whole machine task, and the resource type is a computing device or a computing card; finally, the target scheduling strategy is adopted to adjust the working state of the target computing resource. In this way, when it is detected that there is a target computing resource in the computing system whose temperature exceeds a preset temperature threshold, according to the task type of the target computing task carried by the target computing resource and the resource type of the target computing resource, a task scheduling strategy for reducing the temperature of the target computing resource, i.e., a target scheduling strategy, is determined, and then the working state of the target computing resource is adjusted using the task scheduling strategy, thereby reducing the temperature of the target computing resource and avoiding serious negative effects caused by excessively high temperature of the target computing resource. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A schematic diagram of an application scenario of a computing resource scheduling method provided in an embodiment of the present application;

[0024] Figure 2 A method flow chart of a computing resource scheduling method provided in an embodiment of the present application;

[0025] Figure 3 A schematic diagram of the relationship between the utilization rate and temperature of a computing card provided in an embodiment of the present application;

[0026] Figure 4 A schematic diagram of the relationship between the utilization rate and temperature of a computing device provided in an embodiment of the present application;

[0027] Figure 5 A flowchart of the overall implementation process of a computing resource scheduling method provided in an embodiment of the present application;

[0028] Figure 6 A schematic diagram of a flow chart of a scheduling strategy execution process provided in an embodiment of the present application;

[0029] Figure 7 A schematic diagram of the structure of a computing resource scheduling device provided in an embodiment of the present application;

[0030] Figure 8 is a schematic diagram of the structure of a terminal device provided in an embodiment of the present application;

[0031] Fig. 9A schematic diagram of the structure of a server provided in an embodiment of the present application. DETAILED DESCRIPTION

[0032] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0033] The terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatuses.

[0034] In related technologies, in order to ensure the stable operation of computing devices, the status of the training tasks running on the computing devices is usually fed back, and whether the computing devices are faulty is determined based on the fed-back status of the training tasks; then the resources of the computing devices are scheduled based on the status of the training tasks and the fault conditions of the computing devices. The impact of temperature on the operation of computing devices is not considered, and if the temperature of the computing device or the computing card in the computing device exceeds the temperature critical line, the corresponding computing device will be restarted, and the restart of the computing device will cause the model training tasks it carries to be terminated.

[0035] In view of this, an embodiment of the present application provides a computing resource scheduling method, in which the temperature of each computing resource in the computing system is first obtained; wherein the computing resources include computing devices in the computing system, and computing cards in the computing devices; when it is detected that there is a target computing resource in the computing system whose temperature exceeds a preset temperature threshold, the target computing task carried by the target computing resource is determined; then, according to the task type of the target computing task and the resource type of the target computing resource, a target scheduling strategy is determined; wherein the task type is a whole machine task or a non-whole machine task, and the resource type is a computing device or a computing card; finally, the target scheduling strategy is adopted to adjust the working state of the target computing resource.

[0036] In the above-mentioned computing resource scheduling method, when it is detected that there is a target computing resource in the computing system whose temperature exceeds a preset temperature threshold, a task scheduling strategy for reducing the temperature of the target computing resource, i.e., a target scheduling strategy, is determined based on the task type of the target computing task carried by the target computing resource and the resource type of the target computing resource. The task scheduling strategy is then used to adjust the working state of the target computing resource, thereby reducing the temperature of the target computing resource and avoiding serious negative effects caused by excessively high temperature of the target computing resource.

[0037] The computing resource scheduling method provided in the embodiment of the present application can be executed by a computer device with data processing capabilities, and the computer device can be a terminal device or a server. Among them, the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The terminal device can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited to this. The terminal device and the server can be directly or indirectly connected via wired or wireless communication, and this application is not limited here.

[0038] In order to facilitate understanding of the computing resource scheduling method provided in the embodiment of the present application, the application scenario of the computing resource scheduling method is first exemplarily introduced below.

[0039] See also Figure 1 , Figure 1 A schematic diagram of an application scenario of a computing resource scheduling method provided in an embodiment of the present application. Figure 1 As shown, the application scenario includes a computing system composed of multiple computing devices (computing device 101, computing device 102, computing device 103, ..., computing device 10n). Exemplarily, the following is an example of the computing device 101 in the computing system executing the computing resource scheduling method provided in the embodiment of the present application. It should be understood that in actual applications, the computing resource scheduling method provided in the embodiment of the present application can also be executed by other computer devices outside the computing system, and the embodiment of the present application does not make any limitation to this.

[0040] Each computing device in the computing system can carry at least one computing card, which is used to execute computing tasks (such as model training tasks, etc.) on the computing device. The computing card carried by the computing device can be, for example, a graphics processing unit (GPU) card. In the embodiment of the present application, the computing device in the computing system and the computing card carried by the computing device are regarded as computing resources.

[0041] In specific implementation, computing device 101 can obtain the temperature of each computing resource in the computing system, that is, obtain the temperature of each computing device 101, computing device 102, computing device 103, ..., computing device 10n in the computing system, and the temperature of each computing card carried by computing device 101, computing device 102, computing device 103, ..., computing device 10n.

[0042] Then, the computing device 101 can detect whether there is a temperature exceeding a preset temperature threshold based on the temperatures of each computing resource obtained. If so, the computing resource whose temperature exceeds the preset temperature threshold is determined as a target computing resource, and the computing task carried by the target computing resource is determined as the target computing task.

[0043] Furthermore, the computing device 101 can determine a target scheduling strategy for reducing the temperature of the target computing resource according to the task type of the target computing device (whole machine task or non-whole machine task) and the resource type of the target computing resource (computing device or computing card). Furthermore, the target scheduling strategy is used to adjust the working state of the target computing resource to achieve the goal of lowering the temperature of the target computing resource, thereby avoiding the occurrence of situations such as restarting the computing device due to the excessive temperature of the computing resource, that is, avoiding serious negative effects caused by the excessive temperature of the computing resource.

[0044] It should be understood that Figure 1 The application scenarios shown are only examples. In actual applications, the computing resource scheduling method provided in the embodiments of the present application can also be applied to other application scenarios, and the embodiments of the present application do not impose any limitations on this.

[0045] The computing resource scheduling method provided by this application is introduced below in conjunction with specific embodiments.

[0046] See also Figure 2 , Figure 2 A method flow chart of a computing resource scheduling method provided in an embodiment of the present application. The method comprises the following steps:

[0047] S201, obtaining the temperature of each computing resource in the computing system; the computing resources include computing devices in the computing system and computing cards in the computing devices.

[0048] A computing system is a system for performing computing tasks. The computing system includes multiple computing devices, each of which carries at least one computing card. In the embodiment of the present application, both computing devices and computing cards can be regarded as computing resources. Generally, a computing device can carry multiple computing cards of the same model. For example, the computing device can include 8 GPU cards of the same model. Of course, the computing device can also carry other types of computing cards, and the embodiment of the present application does not impose any limitation on this.

[0049] In the process of computing tasks being performed by the computing system, as the computing tasks proceed, the utilization rate of the computing cards in the computing devices continues to increase, the temperature of the computing cards will continue to increase, and the temperature of the computing devices will continue to increase. In order to prevent the computing tasks being performed from being seriously negatively affected by the excessive temperature of the computing cards or computing devices, it is necessary to obtain the temperature of each computing device in the computing system and the temperature of the computing cards in the computing devices, detect whether there is a temperature exceeding a preset temperature threshold, and decide whether to adopt the relevant resource scheduling strategy based on the detection results.

[0050] S202: When it is detected that there is a target computing resource in the computing system whose temperature exceeds a preset temperature threshold, determine a target computing task carried by the target computing resource.

[0051] According to the acquired temperatures of each computing device in the computing system and each computing card in the computing device, determine whether there is a computing resource whose temperature exceeds a preset temperature threshold. The preset temperature threshold can be set according to actual conditions. For example, the preset temperature threshold can be set to 80° C. The computing resource whose temperature exceeds the preset temperature threshold is determined as the target computing resource.

[0052] When it is detected that there is a target computing resource in the computing system whose temperature exceeds a preset temperature threshold, a computing task carried by the target computing resource is determined as a target computing task, so as to determine the type of the target computing task according to the target computing task.

[0053] S203. Determine a target scheduling strategy according to the task type of the target computing task and the resource type of the target computing resource; the task type is a whole-machine task or a non-whole-machine task, and the resource type is a computing device or a computing card.

[0054] Specifically, the task type can be divided into a whole machine task or a non-whole machine task, and the resource type can be divided into a computing device or a computing card. Determining the type of the target computing task according to the target computing task can be understood as determining whether the target computing task is a whole machine task or a non-whole machine task according to the target computing task.

[0055] It is understandable that a computing task can be executed by at least one computing card. If the computing cards that execute the computing task do not completely cover all the computing cards of a computing device, the task type of the computing task is a non-whole-machine task. For example, assuming that a computing device can carry 8 computing cards, if the number of computing cards used to execute a computing task is less than 8, the computing task can be regarded as a non-whole-machine task. Alternatively, if the number of computing cards used to execute a computing task is more than or equal to 8, but the computing cards used to execute the computing task are deployed on different computing devices, and the number of computing cards used to execute the computing task on each computing device is less than 8, the computing task can also be regarded as a non-whole-machine task. On the contrary, if the computing cards that execute the computing task completely cover all the computing cards of a computing device, the task type of the computing task is a whole-machine task. For example, assuming that a computing device can carry 8 computing cards, if the number of computing cards used to execute a computing task is equal to 8, and the 8 computing cards used to execute the computing task belong to the same computing device, the computing task can be regarded as a whole-machine task.

[0056] After determining the task type of the target computing task, a target scheduling strategy is determined according to the task type of the target computing task and the resource type of the target computing resource. The target scheduling strategy is a task scheduling strategy for reducing the temperature of the target computing resource.

[0057] S204: Adopt a target scheduling strategy to adjust the working status of the target computing resources.

[0058] The target scheduling strategy is used to adjust the working state of the target computing resources to reduce the temperature of the target computing resources, thereby avoiding serious negative impacts on the target computing resources due to excessive temperature.

[0059] The embodiment of the present application provides a computing resource scheduling method, in which the temperature of each computing resource in the computing system is first obtained; wherein the computing resources include computing devices in the computing system and computing cards in the computing devices; when it is detected that there is a target computing resource in the computing system whose temperature exceeds a preset temperature threshold, the target computing task carried by the target computing resource is determined; then the target scheduling strategy is determined according to the task type of the target computing task and the resource type of the target computing resource; wherein the task type is a whole machine task or a non-whole machine task, and the resource type is a computing device or a computing card; finally, the target scheduling strategy is adopted to adjust the working state of the target computing resource. In this way, when it is detected that there is a target computing resource in the computing system whose temperature exceeds a preset temperature threshold, according to the task type of the target computing task carried by the target computing resource and the resource type of the target computing resource, a task scheduling strategy for reducing the temperature of the target computing resource, i.e., a target scheduling strategy, is determined, and then the working state of the target computing resource is adjusted using the task scheduling strategy, thereby reducing the temperature of the target computing resource and avoiding serious negative effects caused by excessively high temperature of the target computing resource.

[0060] In one possible implementation, when the target computing task is a non-whole machine task and the resource type of the target computing resource is a computing card, it can be understood that when an alarm is triggered by a computing card that the temperature exceeds a preset temperature threshold and the computing task carried by the computing card is a non-whole machine task, the target scheduling strategy determined based on the task type of the target computing task and the resource type of the target computing resource is the computing card frequency reduction strategy; the computing card frequency reduction strategy here is used to indicate reducing the operating frequency of the computing card.

[0061] Specifically, adopting the target scheduling strategy to adjust the working state of the target computing resource may include: adopting the computing card frequency reduction strategy to reduce the working frequency of the target computing resource until the temperature of the target computing resource is lower than the preset temperature threshold. It should be understood that reducing the working frequency of the target computing resource can be understood as reducing the clock frequency of the target computing resource, that is, reducing the workload executed by the target computing resource per unit time, reducing the utilization rate of the target computing resource, and thus reducing the temperature of the target computing resource.

[0062] In an embodiment of the present application, when an alarm is triggered by a computing card that the temperature exceeds a preset temperature threshold, and the computing task carried by the computing card is a non-whole machine task, the temperature of the computing card is reduced by reducing the operating frequency of the computing card, thereby avoiding serious negative effects caused by the excessive temperature of the computing card.

[0063] In a possible implementation, when the target computing task is a non-whole machine task and the resource type of the target computing resource is a computing card, the computing card that triggers the temperature alarm may be one computing card or multiple computing cards. It should be understood that each computing card that triggers the temperature alarm here may specifically include one computing card of the card type, or may include multiple computing cards of the card type.

[0064] When only one type of computing card initiates a temperature alarm, the operating frequency of this type of computing card can be directly reduced until the temperature of this type of computing card is lower than a preset temperature threshold.

[0065] When there are multiple types of computing cards that initiate temperature alarms, the target card type can be determined based on the single card temperature models corresponding to the multiple card types. Specifically, the card type whose temperature is reduced by the largest magnitude after the preset utilization rate of the computing card is reduced is determined as the target card type. After determining the target card type, the operating frequency of the computing card corresponding to the target card type is reduced until the temperatures of the computing cards of the multiple card types that are the target computing resources are all lower than the preset temperature threshold. In this way, the target card type whose temperature is reduced by the largest magnitude after the frequency reduction process can be selected from the computing cards of the multiple card types that initiate temperature alarms, and the operating frequency of the computing cards of the target card type can be reduced. In this way, the temperature of the computing card that is the target computing resource can be reduced more efficiently to avoid serious negative effects caused by excessively high temperatures of the target computing resources.

[0066] It should be noted that the single-card temperature model is used to characterize the relationship between the utilization and temperature of the computing card under its corresponding card type. For example, taking the computing card as a GPU computing card as an example, the card types of computing cards include M40 series, P40 series, V100 series, A series and H series. The corresponding relationship between the utilization and temperature of the GPU computing card is specifically introduced below in conjunction with the accompanying drawings.

[0067] See also Figure 3 , Figure 3 A schematic diagram of the relationship between the utilization rate and temperature of a computing card provided in an embodiment of the present application. Figure 3 The horizontal axis is the utilization of the GPU computing card, and the vertical axis is the temperature. Figure 3 It can be seen that as the utilization rate of the GPU computing card increases, the temperature of the GPU computing card will also rise, and after reaching 90% utilization, the temperature change tends to be gentle.

[0068] Specifically, in the embodiment of the present application, the utilization and temperature test can be performed in a single card and single machine granularity manner, and then the temperature corresponding to the utilization of the test can be plotted into a trend chart. Exemplarily, the test of a single card is performed using a single card benchmark tool; the benchmark tool is a program that performs floating-point calculations in a loop, and the process of gradually increasing the pressure on the GPU computing card can be achieved by controlling the number of concurrent loops. By collecting the utilization and temperature data after the pressure is increased, a schematic diagram of the relationship between the utilization and temperature of the GPU computing card is drawn.

[0069] In a possible implementation, the target card type is determined according to the single card temperature models corresponding to the multiple card types, which can be implemented in the following way:

[0070] For each of the multiple card types, according to the current utilization of the computing card of the card type and the single card temperature model corresponding to the card type, the temperature reduction of the computing card of the card type when the current utilization is reduced by a preset amount is determined as the temperature reduction corresponding to the card type. The card type with the largest corresponding temperature reduction among the multiple card types is determined as the target card type.

[0071] Exemplarily, if the computing cards that initiate the temperature alarm include computing cards of type A and type B, the current utilization rate of computing cards of type A is 70%, and the current utilization rate of computing cards of type B is 60%. Taking the preset amplitude of 10% as an example, after the utilization rate of computing cards of type A is reduced by 10%, the temperature of computing cards of type A is reduced by 10°C; and after the utilization rate of computing cards of type B is reduced by 10%, the temperature of computing cards of type B is reduced by 20°C. It can be seen from this that after the utilization rates of computing cards of type A and type B are reduced by the same preset amplitude, the temperature drop of computing cards of type B is greater than that of computing cards of type A, and computing cards of type B are determined as the target card type. After determining the target card type, the operating frequency of computing cards of the target card type is reduced until the temperatures of computing cards of multiple card types are all lower than the preset temperature threshold.

[0072] In an embodiment of the present application, by determining the target card type and lowering the temperature of the target card type, the temperature of the computing card serving as the target computing resource can be more efficiently lowered to avoid serious negative effects caused by excessive temperature of the target computing resource.

[0073] The above embodiment introduces the specific implementation process of reducing the temperature of the computing card in the application scenario where the computing task is not a whole machine task and the temperature alarm is initiated by the computing card. The following describes the specific implementation process of reducing the temperature of the computing device in the application scenario where the computing task is not a whole machine task and the temperature alarm is initiated by the computing device.

[0074] In a possible implementation, when the target computing task is a non-whole machine task and the resource type of the target computing resource is a computing device, it can be understood that the alarm of the temperature exceeding the preset temperature threshold is triggered by the computing device, and the computing task carried by the computing device is a non-whole machine task. In this case, according to the task type of the target computing task and the resource type of the target computing resource, the target scheduling strategy determined is the task frequency reduction strategy; the task frequency reduction strategy here is used to indicate reducing the working frequency of the computing task carried on the computing device.

[0075] When the target scheduling strategy is the task frequency reduction strategy and the target computing resource is a computing device, the target scheduling strategy is adopted to adjust the working state of the target computing resource. The specific implementation process is as follows:

[0076] Among the tasks running on the computing device as the target computing resource, determine the task with the lowest priority, and determine the task with the lowest priority as the target task; then use the target task frequency reduction strategy to reduce the operating frequency of the target task until the temperature of the target computing resource is lower than the preset temperature threshold. It should be understood that reducing the operating frequency of the target task can be understood as reducing the clock frequency of the target task, that is, reducing the execution amount of the target task per unit time, reducing the utilization rate of the computing device carrying the target task, and thus reducing the temperature of the computing device.

[0077] Specifically, if the temperature of the target computing resource still cannot be reduced to below the preset temperature threshold after reducing the operating frequency of the target task, the execution of the target task may be terminated.

[0078] Furthermore, if after the execution of the target task is terminated, the temperature of the target computing resource cannot be reduced to below the preset temperature threshold, the operating frequency of the target computing resource can be reduced, that is, the operating frequency of the computing device can be reduced. If after reducing the operating frequency of the computing device, the temperature of the computing device still cannot be reduced to below the preset temperature threshold, the operation of reducing the throughput of the computing device can be further performed until the temperature of the computing device is reduced to below the preset temperature threshold.

[0079] In an embodiment of the present application, when the target computing task is a non-whole machine task and the resource type of the target computing resource is a computing device, the temperature of the computing device is reduced by determining the lowest priority task among the tasks running on the computing device and reducing the operating frequency of the task, thereby avoiding serious negative effects caused by excessive temperature of the computing device.

[0080] The above embodiment introduces the specific implementation process of reducing the temperature of computing resources when the computing task is a non-whole machine task. The following describes the specific implementation process of reducing the temperature of computing resources when the computing task is a whole machine task.

[0081] In a possible implementation, when the task type of the target computing task is a whole machine task, and the resource type of the target computing resource is a computing device, it can be understood that when an alarm is triggered by the computing device that the temperature exceeds the preset temperature threshold, and the computing task carried by the computing device is a whole machine task, the specific implementation process of determining the target scheduling strategy according to the task type of the target computing task and the resource type of the target computing resource is as follows:

[0082] Determine the proportion of the computing device that is the target computing resource in the various computing devices that carry the target computing task. It should be understood that there can be multiple computing devices that carry the target computing task. Assume that there are three computing devices that carry the target computing task, namely computing device X, computing device Y and computing device Z, where computing device X is the target computing resource. If only computing device X triggers an alarm that the temperature exceeds the preset temperature threshold, then the proportion of the computing device that is the target computing resource in the various computing devices that carry the target computing task is 1 / 3; if both computing device X and computing device Y trigger an alarm that the temperature exceeds the preset temperature threshold, then the proportion of the computing device that is the target computing resource in the various computing devices that carry the target computing task is 2 / 3. Exemplarily, the preset ratio threshold can be set according to actual conditions. In the embodiment of the present application, the preset ratio threshold can be set to 50%.

[0083] If the proportion of the computing device as the target computing resource in the various computing devices carrying the target computing task does not exceed the preset ratio threshold, the target scheduling strategy determined is the computing device frequency reduction strategy. For example, assuming that the determined proportion is 1 / 3, which does not exceed 50%, the operating frequency of computing device X is reduced according to the computing device frequency reduction strategy.

[0084] If the proportion of the computing device as the target computing resource in the various computing devices carrying the target computing task exceeds the preset ratio threshold, the target scheduling strategy determined is the task throughput reduction strategy. Exemplarily, if both computing device X and computing device Y have issued an alarm that the temperature exceeds the preset temperature threshold. For example, assuming that the determined proportion is 2 / 3, which exceeds 50%, the throughput of the target computing task can be reduced according to the task throughput reduction strategy until the temperature of the computing device as the target computing resource drops below the preset temperature threshold.

[0085] In this way, the target scheduling strategy can be determined based on the relationship between the proportion of the computing device serving as the target computing resource in the various computing devices carrying the target computing task and the preset proportion threshold, and the temperature of the computing device serving as the target computing resource can be reduced according to the target scheduling strategy to avoid serious negative effects caused by the computing device serving as the target computing resource being too hot.

[0086] In a possible implementation, when the target scheduling strategy is a computing device frequency reduction strategy, the target scheduling strategy is adopted to adjust the working state of the target computing resource, which can be implemented in the following manner:

[0087] The target computing resource may include one computing device or multiple computing devices. If the target computing resource includes only one computing device, the operating frequency of the computing device as the target computing resource may be reduced until the temperature of the computing device is lower than a preset temperature threshold.

[0088] If the target computing resource includes multiple computing devices, the target computing device can be determined according to the whole machine temperature model corresponding to the card types carried by the multiple computing devices. The target computing device can be understood as the computing device with the fastest temperature drop among the multiple computing devices that trigger the temperature alarm. After the target computing device is determined, the operating frequency of the target computing device can be reduced until the temperature of the multiple computing devices that initiate the temperature alarm is lower than the preset temperature threshold. In this way, by determining the target computing device and reducing the operating frequency of the target computing device, the temperature of the alarm computing device can be reduced more efficiently to avoid the serious negative impact of the computing device that initiates the temperature alarm due to excessive temperature.

[0089] It should be noted that the whole machine temperature model is used to characterize the relationship between the utilization and temperature of the computing device that carries the corresponding card type. It can be understood that, taking the computing device including 8 GPU cards as an example, the utilization of the whole machine can be understood as the cumulative value of the utilization of the 8 GPU cards, and the whole machine temperature model is used to characterize the relationship between the cumulative value of the utilization of the 8 GPU cards in the computing device and the temperature.

[0090] See also Figure 4 , Figure 4 A schematic diagram of the relationship between utilization and temperature of a computing device provided in an embodiment of the present application. Figure 4 The figure shows the relationship between the whole machine utilization and temperature of the computing device, where the horizontal axis is the utilization of the GPU of the whole machine, and the vertical axis is the temperature value of the whole machine. Figure 4 It can be seen that the higher the utilization rate of the GPU of the whole machine, the higher the temperature of the whole machine; after the utilization rate of the whole machine reaches 600%, the temperature change tends to be gentle.

[0091] Exemplarily, for the schematic diagram of the relationship between the utilization and temperature of the GPU whole machine computing device, the embodiment of the present application can use a single machine multi-card benchmark test tool to perform multi-card functional stress testing, and control the whole machine GPU utilization by increasing the amount of floating point calculation data, so as to obtain the whole machine temperature at each utilization, and draw a schematic diagram of the relationship between the utilization and temperature of the GPU whole machine computing device by collecting the utilization and temperature data of the whole machine GPU computing device. It should be understood that for multiple computing devices, according to the method of drawing the relationship between the utilization and temperature of the GPU whole machine computing device, a schematic diagram of the corresponding relationship between the utilization and temperature of each computing device can be drawn respectively.

[0092] In a possible implementation, when the target computing resource includes multiple computing devices, the target computing device is determined according to the whole machine temperature models corresponding to the card types carried by the multiple computing devices, which can be implemented in the following manner:

[0093] For each computing device among the multiple computing devices, according to the whole machine temperature model corresponding to the card type carried by the computing device and the current utilization of the computing device, the temperature drop of the computing device when the current utilization is reduced by a preset amplitude is determined as the temperature drop corresponding to the computing device. Specifically, the relationship between the utilization and temperature of the whole computing device can be regarded as the whole machine temperature model corresponding to the card type carried by the computing device. Exemplarily, the preset amplitude can be set according to the actual situation. In the embodiment of the present application, the preset amplitude can be set to 10%.

[0094] Assume that two computing devices both initiate an alarm that the temperature exceeds the preset temperature threshold, and the two computing devices are computing device X and computing device Y, and computing device X and computing device Y respectively carry computing cards of different card types. Assume that the current utilization rate of computing device X is 60%, and the current utilization rate of computing device Y is 70%; after computing device X reduces its utilization rate by 10%, the temperature of computing device X drops by 15°C; after computing device Y reduces its utilization rate by 10%, the temperature of computing device Y drops by 25°C. According to the temperature drop, determine the computing device with the largest temperature drop among multiple computing devices as the target computing device. It can be seen that the temperature drop of computing device Y is greater than that of computing device X, so computing device Y is used as the target computing device. Reduce the operating frequency of target computing device Y until the temperatures of multiple computing devices are lower than the preset temperature threshold.

[0095] An embodiment of the present application provides a method for determining a target computing device. When multiple computing devices trigger an alarm that the temperature exceeds a preset temperature threshold, the target computing device can be determined and the utilization rate of the target computing device can be reduced to achieve a more efficient and rapid reduction in the temperature of the computing device, thereby avoiding serious negative impacts on the computing device due to excessive temperature.

[0096] In a possible implementation, when the target scheduling strategy is a task throughput reduction strategy, the target scheduling strategy is adopted to adjust the working state of the target computing resources, which can be implemented in the following ways:

[0097] According to the preset data throughput reduction rule, the data throughput of the target computing task is reduced until the temperature of the target computing resource is lower than the preset temperature threshold.

[0098] Exemplarily, the preset data throughput reduction rule can be set to first reduce the data throughput of the target computing task by 10%; if after reducing the data throughput of the target computing task by 10%, the temperature of the target computing resource has not dropped below the preset temperature threshold, the data throughput of the target computing task can be reduced by 20%; if after reducing the data throughput of the target computing task by 20%, the temperature of the target computing resource has not dropped below the preset temperature threshold, the data throughput of the target computing task can be reduced by 40%; if after reducing the data throughput of the target computing task by 40%, the temperature of the target computing resource has not dropped below the preset temperature threshold, the execution of the target computing task can be terminated until the temperature of the target computing resource drops below the preset temperature threshold.

[0099] In this way, when the target computing resource initiation temperature exceeds a preset temperature threshold, the temperature of the target computing resource can be reduced by reducing the throughput of the target computing task to avoid serious negative effects caused by excessive temperature of the target computing resource.

[0100] The above embodiment introduces the specific implementation process of reducing the temperature of the computing device when the computing task is a whole machine task and the temperature alarm is initiated by the computing device. The following describes the specific implementation process of reducing the temperature of the computing card when the computing task is a whole machine task and the temperature alarm is initiated by the computing card.

[0101] In a possible implementation, when a temperature alarm is initiated by a computing card and the computing task carried by the computing card is a whole machine task, the target scheduling strategy is determined to be a computing card frequency reduction strategy according to the task type of the target computing task and the resource type of the target computing resource; when the target scheduling strategy is determined to be the computing card frequency reduction strategy, the target scheduling strategy is adopted to adjust the working state of the target computing resource, which can be implemented in the following manner:

[0102] The computing card frequency reduction strategy is adopted to reduce the operating frequency of the target computing resource until the temperature of the target computing resource is lower than the preset temperature threshold. Generally, the temperature of the computing card can be reduced to below the preset temperature threshold by reducing the operating frequency of the computing card.

[0103] If the temperature of the computing card is not lower than the preset temperature threshold after the operating frequency of the computing card is reduced, the throughput of the computing task can be reduced. Specifically, the data throughput of the computing task can be reduced according to the preset data throughput reduction rules until the temperature of the computing card is lower than the preset temperature threshold.

[0104] In this way, when the computing task is a whole machine task and the computing card initiates a temperature alarm, the temperature of the computing card can be reduced to below the preset temperature threshold by reducing the operating frequency of the computing card, thereby avoiding serious negative effects caused by excessive temperature of the computing card.

[0105] The following takes a computing device carrying a GPU card as an example to introduce the overall implementation process of the computing resource scheduling method provided in an embodiment of the present application.

[0106] See also Figure 5 , Figure 5 The overall implementation process framework diagram of a computing resource scheduling method provided in an embodiment of the present application. The computing resource scheduling method provided in an embodiment of the present application includes three processes. The first process is to collect the utilization rate and corresponding temperature of various card types of each computing device, the utilization rate and temperature of each computing device, and save the collected temperature data to the data center to build a single-card temperature model of the GPU card, and build a whole-machine temperature model of the GPU card. For example, the constructed single-card temperature model of the GPU card can be found in Figure 3 The temperature model of the GPU card can be found in Figure 4 .

[0107] The second process is to collect the utilization and temperature of the computing card of the computing device running the computing task, as well as the utilization and temperature of the entire machine of the computing device running the computing task, and save them to the data center to determine whether there are computing cards or computing devices whose temperatures exceed the preset temperature threshold. Specifically, the utilization and temperature of a single GPU card of a running GPU computing device and the utilization and temperature of the GPU of the entire machine can be reported to the data center. The reported field format is shown in Table 1.

[0108] Table 1

[0109]

[0110]

[0111] In Table 1, each GPU computing card is distinguished by its SN number. It should be understood that the SN number of each GPU card is unique. The data of the single card and the whole machine collected every minute is stored in the data center, and the relationship between the utilization and temperature of the GPU whole machine computing device is obtained according to the data in Table 1.

[0112] The third process is that if the temperature of a computing card or computing device exceeds a preset temperature threshold, the computing device in the computing system determines a scheduling strategy based on the task type of the computing task and the resource type of the computing resource, and uses the scheduling strategy to adjust the working state of the computing resource so that the temperature of the computing resource is reduced to below the preset temperature threshold.

[0113] The following describes the specific implementation process of when a computing resource initiates a temperature alarm, the computing device in the computing system determines the scheduling strategy according to the task type of the computing task and the resource type of the computing resource, and uses the scheduling strategy to adjust the working state of the computing resource.

[0114] See also Figure 6 , Figure 6 A flow chart of a scheduling strategy execution process provided for an embodiment of the present application. When there is a computing resource that triggers a temperature alarm, first determine whether the computing task is a whole-machine task or a non-whole-machine task. If it is a whole-machine task, this type of task usually uses 8 computing cards to perform the computing task. When a temperature alarm occurs, it may be a temperature alarm of the entire computing device, or it may be a temperature alarm initiated by a computing card; if it is a temperature alarm initiated by the entire computing device, it is necessary to perform an operation of reducing the operating frequency of the entire computing device that initiates the alarm. If the proportion of the entire computing device that initiates the alarm in the various computing devices that carry the computing task exceeds the preset ratio threshold, it is necessary to execute an operation to limit the throughput of the computing task. Specifically, the throughput can be limited by first reducing the data throughput of the computing task by 10%; if after reducing the data throughput of the computing task by 10%, the temperature of the entire computing device that initiates the alarm has not dropped below the preset temperature threshold, the data throughput of the computing task can be reduced by 20%; if after reducing the data throughput of the computing task by 20%, the temperature of the entire computing device that initiates the alarm has not dropped below the preset temperature threshold, the data throughput of the computing task can be reduced by 40%; if after reducing the data throughput of the computing task by 40%, the temperature of the entire computing device that issues the alarm has not dropped below the preset temperature threshold, the computing task can be terminated until the temperature of the entire computing device that issues the alarm drops below the preset temperature threshold. If the computing resource that issues the alarm is a computing card, the temperature of the computing card can be reduced to below the preset temperature by reducing the operating frequency of the computing card, without executing the operation of limiting the throughput of the computing task.

[0115] If it is a non-whole machine task, the computing resource of this type of task is a non-whole machine 8-card task, which may be a single machine N cards, or a multi-machine task, but each computing device has less than 8 cards. For non-whole machine tasks, first determine whether the temperature alarm is initiated by the computing card or the computing device; if the temperature alarm is initiated by the computing card and it is a type of computing card, the working frequency of the computing card under the card type can be directly reduced until the temperature of the alarmed computing card drops below the preset temperature threshold. If the computing card that initiates the temperature alarm is a computing card of multiple types, the card type with the largest temperature drop among the multiple card types when the current utilization rate decreases by a preset amplitude can be screened out based on the single card temperature model, and the working frequency of the computing card under this card type can be reduced until the temperature of all computing cards that initiate the temperature alarm is lower than the preset temperature threshold. If the temperature alarm is initiated by the computing device, the low-priority computing tasks on the computing device that initiates the temperature alarm can be screened out, and the computing tasks can be reduced in working frequency, throughput, or even terminated until the temperature of the computing device is lower than the preset temperature threshold.

[0116] The present application also provides a computing resource scheduling device. Figure 7 , Figure 7 A schematic diagram of the structure of a computing resource scheduling device provided in an embodiment of the present application. The device includes:

[0117] The temperature acquisition module 701 is used to acquire the temperature of each computing resource in the computing system; the computing resources include computing devices in the computing system and computing cards in the computing devices;

[0118] The task determination module 702 is used to determine the target computing task carried by the target computing resource when it is detected that there is a target computing resource with a temperature exceeding a preset temperature threshold in the computing system;

[0119] A strategy determination module 703 is used to determine a target scheduling strategy according to a task type of a target computing task and a resource type of a target computing resource; the task type is a whole machine task or a non-whole machine task, and the resource type is a computing device or a computing card;

[0120] The scheduling module 704 is used to adopt a target scheduling strategy to adjust the working state of the target computing resources.

[0121] The embodiment of the present application provides a computing resource scheduling device for executing a corresponding method. When the temperature of a target computing resource exceeds a preset temperature threshold, the working state of the target computing resource is adjusted according to the task type of the target computing task and the resource type of the target computing resource to reduce the temperature of the target computing resource. Specifically, in this method, the temperature of each computing resource in the computing system can be obtained first; wherein the computing resources include computing devices in the computing system, and computing cards in the computing devices; when it is detected that there is a target computing resource in the computing system whose temperature exceeds the preset temperature threshold, the target computing task carried by the target computing resource is determined; then, according to the task type of the target computing task and the resource type of the target computing resource, the target scheduling strategy is determined; wherein the task type is a whole machine task or a non-whole machine task, and the resource type is a computing device or a computing card; finally, the target scheduling strategy is adopted to adjust the working state of the target computing resource. In this way, when it is detected that there is a target computing resource in the computing system whose temperature exceeds a preset temperature threshold, the target computing task carried by the target computing resource is determined, and the target scheduling strategy is determined according to the task type of the target computing task and the resource type of the target computing resource; and the scheduling strategy is used to adjust its working state, thereby reducing the temperature of the target computing resource and avoiding serious negative effects caused by excessively high temperature.

[0122] In an optional embodiment of the present application, the strategy determination module 703 is specifically used to determine that the target scheduling strategy is a computing card frequency reduction strategy when the task type of the target computing task is a non-whole machine task and the resource type of the target computing resource is a computing card;

[0123] The scheduling module 704 is specifically used to adopt a computing card frequency reduction strategy to reduce the operating frequency of the target computing resource until the temperature of the target computing resource is lower than a preset temperature threshold.

[0124] In an optional embodiment of the present application, the scheduling module 704 is specifically used for:

[0125] When the target computing resources include only computing cards of one card type, reducing the operating frequency of the computing cards of the card type until the temperature of the computing cards of the card type is lower than a preset temperature threshold;

[0126] When the target computing resources include computing cards of multiple card types, the target card type is determined according to the single card temperature models corresponding to the multiple card types; the operating frequency of the computing cards of the target card type is reduced until the temperatures of the computing cards of the multiple card types are all lower than a preset temperature threshold.

[0127] In an optional embodiment of the present application, the single-card temperature model is used to characterize the relationship between the utilization rate and temperature of the computing card of the corresponding card type, and the scheduling module 704 is specifically used to:

[0128] For each card type among the multiple card types, according to the current utilization rate of the computing card of the card type and the single card temperature model corresponding to the card type, determine the temperature reduction of the computing card of the card type when the current utilization rate decreases by a preset amount, as the temperature reduction corresponding to the card type;

[0129] A card type having the largest temperature drop among the multiple card types is determined as the target card type.

[0130] In an optional embodiment of the present application, the strategy determination module 703 is specifically used to determine that the target scheduling strategy is a task frequency reduction strategy when the task type of the target computing task is a non-whole machine task and the resource type of the target computing resource is a computing device;

[0131] The scheduling module 704 is specifically used to determine the task with the lowest corresponding priority as the target task among the tasks running on the computing device as the target computing resource;

[0132] The target task frequency reduction strategy is adopted to reduce the operating frequency of the target task until the temperature of the target computing resource is lower than the preset temperature threshold.

[0133] In an optional embodiment of the present application, the strategy determination module 703 is specifically used to:

[0134] When the task type of the target computing task is a whole-machine task and the resource type of the target computing resource is a computing device, determining the proportion of the computing device serving as the target computing resource in each computing device carrying the target computing task;

[0135] If the proportion does not exceed the preset proportion threshold, the target scheduling strategy is determined to be the computing device frequency reduction strategy; if the proportion exceeds the preset proportion threshold, the target scheduling strategy is determined to be the task throughput reduction strategy.

[0136] In an optional embodiment of the present application, when the target scheduling policy is a computing device frequency reduction policy, the scheduling module 704 is specifically configured to:

[0137] When the target computing resource includes only one computing device, reducing the operating frequency of the computing device as the target computing resource until the temperature of the computing device is lower than a preset temperature threshold;

[0138] When the target computing resources include multiple computing devices, the target computing device is determined according to the whole machine temperature models corresponding to the card types carried by each of the multiple computing devices; the operating frequency of the target computing device is reduced until the temperature of the multiple computing devices is lower than the preset temperature threshold.

[0139] In an optional embodiment of the present application, the whole machine temperature model is used to characterize the relationship between the utilization rate and temperature of the computing device carrying the corresponding card type, and the scheduling module 704 is specifically used to:

[0140] For each computing device among the multiple computing devices, according to the whole machine temperature model corresponding to the card type carried by the computing device and the current utilization of the computing device, determine the temperature reduction of the computing device when the current utilization is reduced by a preset range, as the temperature reduction corresponding to the computing device;

[0141] A computing device having the largest temperature drop among the multiple computing devices is determined as the target computing device.

[0142] In an optional embodiment of the present application, when the target scheduling strategy is a task throughput reduction strategy, the scheduling module 704 is specifically configured to:

[0143] According to the preset data throughput reduction rule, the data throughput of the target computing task is reduced until the temperature of the target computing resource is lower than the preset temperature threshold.

[0144] In an optional embodiment of the present application, the strategy determination module 703 is specifically used to determine that the target scheduling strategy is a computing card frequency reduction strategy when the task type of the target computing task is a whole machine task and the resource type of the target computing resource is a computing card;

[0145] The scheduling module 704 is specifically used to adopt a computing card frequency reduction strategy to reduce the operating frequency of the target computing resource until the temperature of the target computing resource is lower than a preset temperature threshold.

[0146] The embodiment of the present application also provides a computer device, which may specifically be a terminal device or a server. The terminal device and the server provided in the embodiment of the present application will be introduced below from the perspective of hardware entity.

[0147] See also Figure 8 , Figure 8 Schematic diagram of the structure of the terminal device provided in the embodiment of the present application. Figure 8 For the sake of convenience, only the parts related to the embodiments of the present application are shown. For specific technical details not disclosed, please refer to the method part of the embodiments of the present application. The terminal can be any terminal device including a mobile phone, a tablet computer, a personal digital assistant, a point of sales (POS), a car computer, etc., taking the terminal as a computer as an example:

[0148] Figure 8 FIG. 1 is a block diagram showing a partial structure of a computer related to a terminal provided in an embodiment of the present application. Figure 8The computer includes: a radio frequency (RF) circuit 1210, a memory 1220, an input unit 1230 (including a touch panel 1231 and other input devices 1232), a display unit 1240 (including a display panel 1241), a sensor 1250, an audio circuit 1260 (which can be connected to a speaker 1261 and a microphone 1262), a wireless fidelity (WiFi) module 1270, a processor 1280, and a power supply 1290. Those skilled in the art can understand that Figure 8 The computer structure shown in the figure does not constitute a limitation of the computer, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0149] The memory 1220 can be used to store software programs and modules. The processor 1280 executes various functional applications and data processing of the computer by running the software programs and modules stored in the memory 1220. The memory 1220 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the computer (such as audio data, a phone book, etc.), etc. In addition, the memory 1220 can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0150] The processor 1280 is the control center of the computer, which uses various interfaces and lines to connect various parts of the entire computer, and executes various functions of the computer and processes data by running or executing software programs and / or modules stored in the memory 1220, and calling data stored in the memory 1220. Optionally, the processor 1280 may include one or more processing units; preferably, the processor 1280 may integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface, and application programs, and the modem processor mainly processes wireless communications. It is understandable that the above-mentioned modem processor may not be integrated into the processor 1280.

[0151] In the embodiment of the present application, the processor 1280 included in the terminal also has the following functions:

[0152] Obtaining the temperature of each computing resource in the computing system; the computing resources include computing devices in the computing system and computing cards in the computing devices;

[0153] When it is detected that there is a target computing resource in the computing system whose temperature exceeds a preset temperature threshold, determining a target computing task carried by the target computing resource;

[0154] Determine the target scheduling strategy according to the task type of the target computing task and the resource type of the target computing resource; the task type is a whole machine task or a non-whole machine task, and the resource type is a computing device or a computing card;

[0155] Adopt the target scheduling strategy to adjust the working status of the target computing resources.

[0156] Optionally, the processor 1280 is also used to execute the steps of any implementation method of the computing resource scheduling method provided in the embodiments of the present application.

[0157] See also Fig. 9 , Fig. 9 A schematic diagram of the structure of a server provided for an embodiment of the present application. The server 1300 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPU) 1322 (for example, one or more processors) and memory 1332, and one or more storage media 1330 (for example, one or more mass storage devices) storing application programs 1342 or data 1344. Among them, the memory 1332 and the storage medium 1330 may be short-term storage or permanent storage. The program stored in the storage medium 1330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the server. Furthermore, the central processing unit 1322 may be configured to communicate with the storage medium 1330 to execute a series of instruction operations in the storage medium 1330 on the server 1300.

[0158] The server 1300 may also include one or more power supplies 1326, one or more wired or wireless network interfaces 1350, one or more input and output interfaces 1358, and / or one or more operating systems, such as Windows Server 2003. TM , Mac OS X TM , Unix TM ,Linux TM , FreeBSD TM etc.

[0159] The steps performed by the server in the above embodiment can be based on the Fig. 9 The server structure shown.

[0160] The CPU 1322 is used to execute the following steps:

[0161] Obtaining the temperature of each computing resource in the computing system; the computing resources include computing devices in the computing system and computing cards in the computing devices;

[0162] When it is detected that there is a target computing resource in the computing system whose temperature exceeds a preset temperature threshold, determining a target computing task carried by the target computing resource;

[0163] Determine the target scheduling strategy according to the task type of the target computing task and the resource type of the target computing resource; the task type is a whole machine task or a non-whole machine task, and the resource type is a computing device or a computing card;

[0164] Adopt the target scheduling strategy to adjust the working status of the target computing resources.

[0165] Optionally, CPU 1322 may also be used to execute steps of any implementation of the computing resource scheduling method provided in the embodiments of the present application.

[0166] An embodiment of the present application also provides a computer-readable storage medium for storing a computer program, wherein the computer program is used to execute any one of the implementation methods of a computing resource scheduling method described in the aforementioned embodiments.

[0167] The embodiment of the present application also provides a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. The processor of the computer device reads the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes any one of the implementation methods of a computing resource scheduling method described in the above embodiments.

[0168] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0169] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0170] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0171] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0172] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), disk or optical disk and other media that can store computer programs.

[0173] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0174] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A computing resource scheduling method, characterized in that: The method comprises: Acquire the temperature of each computing resource in the computing system; the computing resources include computing devices in the computing system and computing cards in the computing devices; When detecting that there is a target computing resource in the computing system whose temperature exceeds a preset temperature threshold, determining a target computing task carried by the target computing resource; Determine a target scheduling strategy according to the task type of the target computing task and the resource type of the target computing resource; the task type is a whole machine task or a non-whole machine task, and the resource type is a computing device or a computing card; The target scheduling strategy is adopted to adjust the working state of the target computing resource.

2. The method according to claim 1, characterized in that The determining the target scheduling strategy according to the task type of the target computing task and the resource type of the target computing resource includes: When the task type of the target computing task is a non-whole machine task, and the resource type of the target computing resource is a computing card, determining that the target scheduling strategy is a computing card frequency reduction strategy; The adopting the target scheduling strategy to adjust the working state of the target computing resource includes: The computing card frequency reduction strategy is adopted to reduce the operating frequency of the target computing resource until the temperature of the target computing resource is lower than the preset temperature threshold.

3. The method according to claim 2, characterized in that The adopting the computing card frequency reduction strategy to reduce the operating frequency of the target computing resource until the temperature of the target computing resource is lower than the preset temperature threshold includes: When the target computing resource includes only computing cards of one card type, reducing the operating frequency of the computing cards of the card type until the temperature of the computing cards of the card type is lower than the preset temperature threshold; When the target computing resource includes computing cards of multiple card types, the target card type is determined according to the single card temperature models corresponding to the multiple card types; and the operating frequency of the computing cards of the target card type is reduced until the temperatures of the computing cards of the multiple card types are all lower than the preset temperature threshold.

4. The method according to claim 3, characterized in that The single card temperature model is used to characterize the relationship between the utilization rate and temperature of the computing card under the corresponding card type; and determining the target card type according to the single card temperature models corresponding to the multiple card types, comprises: For each card type among the multiple card types, according to the current utilization rate of the computing card of the card type and the single card temperature model corresponding to the card type, determine the temperature reduction of the computing card of the card type when the current utilization rate decreases by a preset amplitude, as the temperature reduction corresponding to the card type; A card type corresponding to the largest temperature drop among the multiple card types is determined as the target card type.

5. The method according to claim 1, characterized in that The determining the target scheduling strategy according to the task type of the target computing task and the resource type of the target computing resource includes: When the task type of the target computing task is a non-whole machine task, and the resource type of the target computing resource is a computing device, determining that the target scheduling strategy is a task frequency reduction strategy; The adopting the target scheduling strategy to adjust the working state of the target computing resource includes: Among the tasks running on the computing device serving as the target computing resource, determining the task with the lowest corresponding priority as the target task; The target task frequency reduction strategy is adopted to reduce the operating frequency of the target task until the temperature of the target computing resource is lower than the preset temperature threshold.

6. The method according to claim 1, characterized in that The determining the target scheduling strategy according to the task type of the target computing task and the resource type of the target computing resource includes: When the task type of the target computing task is a whole-machine task, and the resource type of the target computing resource is a computing device, determining the proportion of the computing device serving as the target computing resource in each computing device carrying the target computing task; If the proportion does not exceed the preset proportion threshold, the target scheduling strategy is determined to be a computing device frequency reduction strategy; if the proportion exceeds the preset proportion threshold, the target scheduling strategy is determined to be a task throughput reduction strategy.

7. The method according to claim 6, characterized in that When the target scheduling strategy is a computing device frequency reduction strategy, the adopting the target scheduling strategy to adjust the working state of the target computing resource includes: When the target computing resource includes only one computing device, reducing the operating frequency of the computing device serving as the target computing resource until the temperature of the computing device is lower than the preset temperature threshold; When the target computing resource includes multiple computing devices, the target computing device is determined according to the whole machine temperature models corresponding to the card types respectively carried by the multiple computing devices; the operating frequency of the target computing device is reduced until the temperature of the multiple computing devices is lower than the preset temperature threshold.

8. The method according to claim 7, characterized in that The whole machine temperature model is used to characterize the relationship between the utilization rate and temperature of the computing device carrying the corresponding card type; determining the target computing device according to the whole machine temperature model corresponding to the card type carried by each of the plurality of computing devices, comprises: For each computing device among the multiple computing devices, according to a whole-machine temperature model corresponding to a card type carried by the computing device and a current utilization rate of the computing device, determining a temperature reduction of the computing device when the current utilization rate decreases by a preset range, as the temperature reduction corresponding to the computing device; A computing device corresponding to the largest temperature drop among the multiple computing devices is determined as the target computing device.

9. The method according to claim 6, characterized in that When the target scheduling strategy is a task throughput reduction strategy, the adopting the target scheduling strategy to adjust the working state of the target computing resource includes: According to a preset data throughput reduction rule, the data throughput of the target computing task is reduced until the temperature of the target computing resource is lower than the preset temperature threshold.

10. The method according to claim 1, characterized in that The determining the target scheduling strategy according to the task type of the target computing task and the resource type of the target computing resource includes: When the task type of the target computing task is a whole machine task, and the resource type of the target computing resource is a computing card, determining that the target scheduling strategy is a computing card frequency reduction strategy; The adopting the target scheduling strategy to adjust the working state of the target computing resource includes: The computing card frequency reduction strategy is adopted to reduce the operating frequency of the target computing resource until the temperature of the target computing resource is lower than the preset temperature threshold.

11. A computing resource scheduling device, characterized in that: The device comprises: A temperature acquisition module, used to acquire the temperature of each computing resource in the computing system; the computing resources include computing devices in the computing system and computing cards in the computing devices; A task determination module, configured to determine a target computing task carried by a target computing resource when it is detected that there is a target computing resource in the computing system whose temperature exceeds a preset temperature threshold; A strategy determination module, used to determine a target scheduling strategy according to a task type of the target computing task and a resource type of the target computing resource; the task type is a whole machine task or a non-whole machine task, and the resource type is a computing device or a computing card; The scheduling module is used to adopt the target scheduling strategy to adjust the working state of the target computing resource.

12. A computer device, characterized in that: The computer device includes a processor and a memory; The memory is used to store computer programs; The processor is configured to execute the computing resource scheduling method according to any one of claims 1 to 10 according to the computer program.

13. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store a computer program, and the computer program is used to execute the computing resource scheduling method according to any one of claims 1 to 10.

Images