Computing power scheduling system and method, and computer equipment
The computing power scheduling system addresses inefficiencies in AI resource management by centralizing resources, dynamically allocating, and integrating platforms, enhancing flexibility and utilization.
Patent Information
- Application Number
- CN202411130850.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-16
- Publication Date
- 2024-12-06
AI Technical Summary
Existing AI computing power scheduling technologies face challenges in managing scattered resources, requiring manual tuning and fixed configurations, leading to inflexible and inefficient resource allocation.
A computing power scheduling system that includes a resource pool module for centralized management, a computing power control module for dynamic allocation, a functional component module for sharing, and a deployment form module for cross-platform integration, enabling flexible and efficient resource utilization.
The system allows for real-time monitoring and dynamic scheduling of computing resources, improving utilization rates and enabling flexible, cross-platform management of computing power.
Smart Images

Figure CN119088552A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of computing power scheduling, and in particular to a computing power scheduling system, method, computer device, computer-readable storage medium, and computer program product. Background Art
[0002] With the development of new-generation information technologies such as big data and cloud computing, computing power scheduling technologies for multiple fields, multiple platforms, and multiple dimensions have emerged. Among them, AI (Artificial Intelligence) computing power refers to the amount of computing required to process a task or realize a function through AI, and it is also a reflection of the computing power of some hardware facilities.
[0003] In traditional technologies, when implementing AI computing power scheduling, resources are scattered and difficult to manage in a unified manner, so additional operation and maintenance and tuning work needs to be deployed. However, this approach that relies on additional operation and maintenance and tuning generally has fixed resource configuration, requires a lot of manual tuning work, and has the problem of insufficient flexibility in resource allocation, and cannot achieve dynamic and rapid resource allocation. Summary of the invention
[0004] Based on this, it is necessary to provide a computing power scheduling system, method, computer equipment, computer-readable storage medium and computer program product that can flexibly and quickly allocate resources and effectively improve computing power utilization in response to the above technical problems.
[0005] In a first aspect, the present application provides a computing power scheduling system, including:
[0006] Resource pool module, computing power control module, functional component module, component communication module, and deployment form module, among which:
[0007] The resource pool module is used to build a computing resource pool, introduce computing resources into the computing resource pool, and classify and centrally manage the introduced computing resources;
[0008] The computing power control module is used to allocate the computing power resources of the computing power resource pool in a fragmented manner;
[0009] The functional component module is established in the computing resource pool architecture constructed by the resource pool module, and is used to complete the computing power sharing between different functional components;
[0010] The component communication module is used to complete the communication between different functional components;
[0011] The deployment form module is used to integrate multiple container cloud platforms to complete cross-platform computing power scheduling.
[0012] In one of the embodiments, the resource pool module is provided with a plurality of resource nodes, and computing resources are introduced into the computing resource pool through the resource nodes; wherein the computing resources include: a physical GPU (Graphics Processing Unit, graphics processing unit), a VGPU (Virtual Graphics Processing Unit, virtual graphics processing unit);
[0013] The resource pool module is specifically used for:
[0014] Decoupling an artificial intelligence AI application from a physical graphics processing unit GPU, and providing a virtual graphics processing unit VGPU for the AI application;
[0015] The computing resources are dynamically scheduled and the status of the computing resources is monitored in real time.
[0016] In one of the embodiments, the resource pool module is provided with a plurality of resource nodes, and computing resources are introduced into the computing resource pool through the resource nodes; wherein the computing resources include: a physical graphics processing unit GPU, a virtual graphics processing unit VGPU;
[0017] The resource pool module is specifically used for:
[0018] Decoupling an artificial intelligence AI application from a physical graphics processing unit GPU, and providing a virtual graphics processing unit VGPU for the AI application;
[0019] The computing resources are dynamically scheduled and the status of the computing resources is monitored in real time.
[0020] In one embodiment, the computing power control module is specifically used to:
[0021] The computing resources introduced by the resource pool module are optimized, and the physical GPU is sliced into VGPUs of any size.
[0022] In one embodiment, the functional component module includes at least one centralized component control core, and the functional component module is specifically used to:
[0023] Connecting each functional component to the component control core via a network and maintaining information synchronization between the functional components;
[0024] Aggregate various information of resource nodes to the component control core, wherein the information includes: any one or more of Internet Protocol IP addresses, physical GPU information, virtual GPU information, and application task information;
[0025] The component control core is used to provide any one or more of the following functions:
[0026] Service registration and service discovery functions of each distributed functional component;
[0027] Flexible VGPU scheduling and allocation function;
[0028] Metadata storage and management of multiple copies;
[0029] License management;
[0030] Provides various presentation layer state transfer application programming interfaces (RestAPI) required for operation and maintenance.
[0031] In one embodiment, the component communication module is specifically used to:
[0032] The communication between various functional components is established through the management plane network and the data plane network, and the resource pool module is cooperated to complete the management of the computing resource pool and the scheduling of computing resources.
[0033] In one embodiment, the deployment form module is specifically used to deploy computing resources on multiple container cloud platforms respectively, wherein the deployment mode includes any one of the following:
[0034] After installing the operating system, deploy each functional component of the computing resource pool directly in binary form;
[0035] Deploy each functional component of the computing resource pool in a containerized manner.
[0036] In one embodiment, it further includes: a management plane establishment module,
[0037] The management plane establishment module is used to establish a management plane through the transmission control protocol / network protocol TCP / IP when deploying the computing power resource pool, so as to complete the visual management of the computing power.
[0038] In a second aspect, the present application further provides a computing power scheduling method, which is applied to the computing power scheduling system in the first aspect, and the method includes:
[0039] Concentrate computing resources from different sources into a computing resource pool, wherein the computing resources include: physical GPUs;
[0040] Slice the physical GPU into VGPUs of any size and dynamically monitor the status of each VGPU in real time;
[0041] When part of the computing power resources need to be called to perform the target task, the target task is systematically tapped, the computing power value required to complete the target task is estimated, and the scope is expanded according to the carrier of the adapted operation to allocate sufficient computing power to the target task;
[0042] After the computing power value estimation is completed, the corresponding amount of computing power resources are directly called through the computing power resource pool to be allocated to the target task.
[0043] In one embodiment, the method further comprises:
[0044] When multiple target tasks are being carried out simultaneously, cross-resource configuration is performed, with the start and end time of computing resources as the configuration division rule, and adaptive configuration of the same area is performed;
[0045] During the process of computing resource allocation and use, the status of computing resources is monitored dynamically in real time;
[0046] After the target task is completed, the corresponding computing resources are automatically returned to the computing resource pool to wait for secondary configuration.
[0047] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:
[0048] Concentrate computing resources from different sources into a computing resource pool, wherein the computing resources include: physical GPUs;
[0049] Slice the physical GPU into VGPUs of any size and dynamically monitor the status of each VGPU in real time;
[0050] When part of the computing power resources need to be called to perform the target task, the target task is systematically tapped, the computing power value required to complete the target task is estimated, and the scope is expanded according to the carrier of the adapted operation to allocate sufficient computing power to the target task;
[0051] After the computing power value estimation is completed, the corresponding amount of computing power resources are directly called through the computing power resource pool to be allocated to the target task.
[0052] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:
[0053] Concentrate computing resources from different sources into a computing resource pool, wherein the computing resources include: physical GPUs;
[0054] Slice the physical GPU into VGPUs of any size and dynamically monitor the status of each VGPU in real time;
[0055] When part of the computing power resources need to be called to perform the target task, the target task is systematically tapped, the computing power value required to complete the target task is estimated, and the scope is expanded according to the carrier of the adapted operation to allocate sufficient computing power to the target task;
[0056] After the computing power value estimation is completed, the corresponding amount of computing power resources are directly called through the computing power resource pool to be allocated to the target task.
[0057] In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor:
[0058] Concentrate computing resources from different sources into a computing resource pool, wherein the computing resources include: physical GPUs;
[0059] Slice the physical GPU into VGPUs of any size and dynamically monitor the status of each VGPU in real time;
[0060] When part of the computing power resources need to be called to perform the target task, the target task is systematically tapped, the computing power value required to complete the target task is estimated, and the scope is expanded according to the carrier of the adapted operation to allocate sufficient computing power to the target task;
[0061] After the computing power value estimation is completed, the corresponding amount of computing power resources are directly called through the computing power resource pool to be allocated to the target task.
[0062] The above-mentioned computing power scheduling system, method, computer device, computer-readable storage medium and computer program product, in which the system includes: a resource pool module, a computing power control module, a functional component module, a component communication module, and a deployment form module. The computing power resource pool is constructed through the resource pool module, and computing power resources are introduced into the computing power resource pool, and the introduced computing power resources are classified and centrally managed; thereby realizing unified resource management, monitoring, scheduling and recycling functions. The computing power resources of the computing power resource pool are fragmented and allocated through the computing power control module, so that the status of each computing power resource in the computing power resource pool can be monitored to improve the utilization rate of computing power resources. By establishing a functional component module in the computing power resource pool architecture constructed by the resource pool module, it is used to complete the computing power sharing between different functional components, so that the computing power can be balanced for the realization of various functions. Communication between different functional components is completed through the component communication module to achieve information synchronization between different functional components. By integrating multiple container cloud platforms through the deployment form module, cross-platform computing power scheduling is completed, so that the corresponding management plane can be established on the system surface to achieve visual management of computing power, and through the deployment form, it can be integrated with multiple container cloud platforms to achieve cross-platform, cross-domain, and multi-dimensional comprehensive management of computing power. The above system can flexibly supervise computing power during computing power scheduling, so that computing power resources can be rationally applied, computing power resources can be shared, and computing power utilization can be significantly improved to achieve the purpose of computing power scheduling. BRIEF DESCRIPTION OF THE DRAWINGS
[0063] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.
[0064] FIG1 is a diagram of an application environment of a computing power scheduling system in one embodiment;
[0065] FIG2 is a structural block diagram of a computing power scheduling system in one embodiment;
[0066] FIG3 is a structural block diagram of a computing power scheduling system in another embodiment;
[0067] FIG4 is a schematic diagram of a flow chart of a computing power scheduling method in one embodiment;
[0068] FIG5 is a schematic diagram of a flow chart of a computing power scheduling method in another embodiment;
[0069] Figure 6 is a structural block diagram of a computing power scheduling device in one embodiment;
[0070] FIG. 7 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION
[0071] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0072] The computing power scheduling system provided in the embodiment of the present application can be applied to the application environment shown in Figure 1. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. Exemplarily, the computing power scheduling system can be set on the server 104, and the GPU (Graphics Processing Unit) in the terminal 102 that is connected to the server 104 can be used as one or more computing power resources by the computing power scheduling system in the server 104. First, a computing power resource pool is constructed, and computing power resources are introduced into the computing power resource pool, and the introduced computing power resources are classified and centrally managed; the computing power resources of the computing power resource pool are fragmented and allocated; computing power sharing and communication between different functional components in the computing power resource pool architecture are established, and when deployed, multiple container cloud platforms are integrated to complete cross-platform computing power scheduling. The terminal 102 may be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, IoT devices, and portable wearable devices. The IoT devices may be smart speakers, smart TVs, smart air conditioners, smart car devices, projection devices, etc. Portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc. Head-mounted devices may be virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, etc. The server 104 may be an independent physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides cloud computing services.
[0073] In an exemplary embodiment, as shown in FIG2 , a computing power scheduling system is provided, and the computing power scheduling system may include: a resource pool module 201, a computing power control module 202, a functional component module 203, a component communication module 204, and a deployment form module 205, wherein: the resource pool module 201 is used to construct a computing power resource pool, introduce computing power resources into the computing power resource pool, and classify and centrally manage the introduced computing power resources; the computing power control module 202 is used to fragment the computing power resources of the computing power resource pool; the functional component module 203 is established in the computing power resource pool architecture constructed by the resource pool module 201, and is used to complete the computing power sharing between different functional components; the component communication module 204 is used to complete the communication between different functional components; the deployment form module 205 is used to integrate multiple container cloud platforms to complete cross-platform computing power scheduling.
[0074] In this embodiment, a resource pool architecture can be constructed through the resource pool module 201, and then computing resources can be introduced into the resource pool by setting multiple resource nodes in the resource pool architecture. All introduced computing resources are classified and centrally managed through the resource pool, thereby realizing remote calling, flexible scheduling and separate deployment. In addition, real-time status monitoring can be performed on all computing resources at the same time to ensure that computing power can be called in real time.
[0075] Exemplarily, the resource pool module 201 is provided with a plurality of resource nodes, and computing power resources are introduced into the computing power resource pool through the resource nodes; wherein the computing power resources include: physical GPU, VGPU (Virtual Graphics Processing Unit); the resource pool module is specifically used for: decoupling artificial intelligence AI applications and physical GPUs, and providing VGPU for the AI applications; dynamically scheduling the computing power resources, and monitoring the status of the computing power resources in real time.
[0076] In this embodiment, a unified physical GPU resource scheduling management and monitoring platform can be constructed through the resource pool module 201, so as to realize functions such as resource management, monitoring, scheduling and recycling. By decoupling AI applications and physical GPUs, VGPU is provided for AI applications, and VGPU resources are scheduled and dynamically scaled on demand, so as to realize on-demand allocation of VGPU resources and on-demand use.
[0077] Exemplarily, the computing power control module 202 is specifically used to optimize the computing power resources introduced by the resource pool module and slice the physical GPU into VGPUs of any size.
[0078] In this embodiment, the performance of the entire computing power scheduling system can be optimized through the computing power control module 202, thereby minimizing resource consumption and ensuring that the performance of the VGPU is similar to that of the physical GPU. By slicing the physical GPU into VGPUs of any size and then scheduling the VGPU to multiple target tasks, multiple AI loads can be allowed to run in parallel, significantly improving the utilization of the physical GPU, thereby ensuring that all computing power resources can be allocated and used in a fragmented manner, so that the input computing power resources can be used in one-to-many or many-to-many scenarios.
[0079] Exemplarily, the functional component module 203 includes at least one centralized component control core, and the functional component module 203 is specifically used to: connect each functional component to the component control core through a network and maintain information synchronization between each functional component; aggregate various information of resource nodes to the component control core, wherein the information includes: any one or more of Internet Protocol IP address, physical GPU information, VGPU information, and application task information.
[0080] In this embodiment, a centralized component control core can be established, which is the core management and scheduling module of the GPU resource pool. All other functional components are directly or indirectly connected to the component control core through the network and keep information synchronization with it. In order to achieve unified management and resource scheduling of the resource pool, the node IP address, physical GPU information, VGPU information, and application task information are all aggregated to the component control core.
[0081] For example, generally, only one component control core may be deployed in one resource pool. However, in order to improve the reliability of the component control core, 2+1 redundancy backup may also be performed, that is, multiple component control cores may be set.
[0082] In this embodiment, the component control core is used to provide any one or more of the following functions:
[0083] Service registration and service discovery functions of each distributed functional component;
[0084] Flexible VGPU scheduling and allocation function;
[0085] Multi-copy highly available metadata storage and management;
[0086] License management;
[0087] Provides various presentation layer state transfer application program interfaces RestAPI required for operation and maintenance.
[0088] Exemplarily, the component communication module 204 is specifically used to establish communication between various functional components through the management plane network and the data plane network, and cooperate with the resource pool module to complete the management of the computing resource pool and the scheduling of computing resources.
[0089] In this embodiment, the communication between functional components means that each functional component communicates through the management plane network and the data plane network to jointly complete functions such as the management of the physical GPU resource pool and the scheduling of physical GPU resources.
[0090] Exemplarily, the deployment mode module 205 is specifically used to deploy computing resources on multiple container cloud platforms respectively, wherein the deployment mode includes any of the following:
[0091] After installing the operating system, deploy each functional component of the computing resource pool directly in binary form;
[0092] Deploy each functional component of the computing resource pool in a containerized manner.
[0093] In this embodiment, the deployment form is to deploy the various functional components of the computing resource pool. For example, it supports directly deploying each functional component on a bare metal server, that is, after installing the operating system, it is directly deployed in Binary form. Of course, containerized deployment is also supported. The deployment method in this embodiment enables the system as a whole to adapt to a variety of Linux operating systems and cloud platforms, so that the entire computing scheduling system has a variety of deployment forms.
[0094] The system in this embodiment includes: a resource pool module, a computing power control module, a functional component module, a component communication module, and a deployment form module, wherein: a computing power resource pool is constructed through the resource pool module, computing power resources are introduced into the computing power resource pool, and the introduced computing power resources are classified and centrally managed; thereby realizing unified resource management, monitoring, scheduling and recycling functions. The computing power resources of the computing power resource pool are fragmented and allocated through the computing power control module, so that the status of each computing power resource of the computing power resource pool can be monitored to improve the utilization rate of computing power resources. By establishing a functional component module in the computing power resource pool architecture constructed by the resource pool module, it is used to complete the computing power sharing between different functional components, so that the computing power can be balanced for the realization of various functions. Communication between different functional components is completed through the component communication module to achieve information synchronization between different functional components. Multiple container cloud platforms are integrated through the deployment form module to complete cross-platform computing power scheduling, so that a corresponding management plane can be established on the surface of the system to realize visual management of computing power, and through the deployment form, it is integrated with multiple container cloud platforms to achieve cross-platform, cross-domain, and multi-dimensional comprehensive management of computing power. The above system can flexibly supervise the computing power during computing power scheduling, so that computing power resources can be rationally applied, computing power resources can be shared, and the utilization rate of computing power can be significantly improved, thereby achieving the purpose of computing power scheduling.
[0095] In an exemplary embodiment, as shown in FIG3 , another computing power scheduling system is provided. The system, based on the system shown in FIG2 , may further include: a management plane establishment module 206. The management plane establishment module 206 is used to establish a management plane through the transmission control protocol / network protocol TCP / IP when deploying a computing power resource pool, so as to complete visual management of computing power.
[0096] In an optional implementation, taking the OrionX AI computing resource pool as an example, when the computing resource pool is deployed, a management plane based on the TCP / IP network is used to carry the management of the entire system. Through the management network, the functional components distributed in each node are synchronized with the OrionXController. Among them, through a private synchronization protocol, each functional component of the distributed deployment has the following characteristics:
[0097] OrionXController supports multi-copy and high-availability deployment modes;
[0098] There is no requirement for the order in which the functional components are started;
[0099] When a functional component recovers from an error, it can be automatically synchronized to the correct state.
[0100] In this embodiment, by establishing a management plane module, the entire computing power scheduling system can be managed through the network, and each management process can be displayed visually, so that the staff can monitor the scheduling of computing power resources.
[0101] In an exemplary embodiment, as shown in FIG4 , a computing power scheduling method is provided, which can be applied to the computing power scheduling system shown in FIG2 and FIG3 , and the method may include the following steps:
[0102] Step 401, centralize computing resources from different sources into a computing resource pool.
[0103] In this embodiment, AI computing resources (such as physical GPUs) from different terminals are first managed in a resource pool to facilitate centralized fragmentation management of the physical GPUs in the resource pool.
[0104] Step 402: Slice the physical GPU into VGPUs of any size, and dynamically monitor the status of each VGPU in real time.
[0105] In this embodiment, the physical GPU can be customized and segmented according to the two dimensions of computing power and video memory to obtain a VGPU that meets the needs of AI applications.
[0106] Step 403, when part of the computing power resources need to be called to execute the target task, the target task is systematically tapped to estimate the computing power value required to complete the target task.
[0107] In this embodiment, the scope of the calculation can be expanded according to the carrier of the adapted operation, so that sufficient computing power can be allocated to the target task.
[0108] Step 404, after the computing power value estimation is completed, the corresponding amount of computing power resources are directly called through the computing power resource pool to be allocated to the target task.
[0109] In this embodiment, according to the estimated computing power value, the corresponding number of computing resources are directly called through the computing power resource pool to be allocated to the target task, that is, the corresponding number of VGPUs are called to execute the target task, where the VGPU can be AI computing power from different terminals.
[0110] In this embodiment, a computing resource pool is established, and the physical GPU is arbitrarily fragmented to obtain a VGPU, and then the VGPU is uniformly scheduled to achieve the management, monitoring, scheduling and recycling of AI computing resources. In addition, by monitoring the computing resource pool, the purpose of computing power control can be achieved, and when the computing power is scheduled, the real-time status of the computing resources is supervised, so that the computing resources can be rationally applied, computing resources can be shared, and the utilization rate of computing power can be improved.
[0111] In an exemplary embodiment, as shown in FIG5 , another computing power scheduling method is provided, which can be applied to the computing power scheduling system shown in FIG2 and FIG3 . The method may include the following steps:
[0112] Step 501, centralize computing resources from different sources into a computing resource pool.
[0113] Step 502: Slice the physical GPU into VGPUs of any size, and dynamically monitor the status of each VGPU in real time.
[0114] Step 503, when part of the computing power resources need to be called to execute the target task, the target task is systematically tapped to estimate the computing power value required to complete the target task.
[0115] Step 504, after the computing power value estimation is completed, the corresponding amount of computing power resources are directly called through the computing power resource pool to be allocated to the target task.
[0116] For the specific implementation process and technical effects of steps 501 to 504 in this embodiment, please refer to the relevant descriptions of steps 401 to 404 in the method shown in FIG. 4 , which will not be repeated here.
[0117] Step 505, when there are multiple target tasks being carried out simultaneously, cross-resource configuration is performed, and the start and end time of the operation of the computing resources is used as the configuration division rule to perform adaptive configuration of the same area.
[0118] In this embodiment, if there are multiple target tasks being carried out at the same time, the start and end time of each computing resource can be obtained, and when the computing resource executing target task A ends, it is promptly allocated to target task B. Alternatively, if the computing resource pre-allocated to target task A has not started running, the computing resource allocated to target task A can be preferentially allocated to target task B to achieve cross-allocation of computing resources for multiple tasks.
[0119] Step 506: During the process of allocating and using computing resources, the status of computing resources is dynamically monitored in real time.
[0120] In this embodiment, as can be seen from step 506, the start and end time of each computing resource can be obtained, so the computing resources occupied by the currently executed target task can be monitored, as well as when the target task releases the occupied computing resources. When multiple target tasks are executed simultaneously or sequentially (with execution time differences), a timetable for the computing resources occupied by each target task can also be set to dynamically adjust the allocated computing resources.
[0121] Step 507: After the target task is completed, the corresponding computing resources are automatically returned to the computing resource pool to wait for secondary configuration.
[0122] In this embodiment, when the target task is completed, the computing resources occupied by the target task are automatically returned to the computing resource pool, and the recovered computing resources can be used for secondary configuration. This enables flexible computing resource scheduling, recovery, etc., and significantly improves the utilization rate of each computing resource.
[0123] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.
[0124] Based on the same inventive concept, the embodiment of the present application also provides a computing power scheduling device for implementing the computing power scheduling method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more computing power scheduling device embodiments provided below can refer to the limitations of the computing power scheduling method above, and will not be repeated here.
[0125] In an exemplary embodiment, Figure 6 As shown, a computing power scheduling device is provided, including: a computing power concentration unit 601, a computing power division and monitoring unit 602, a computing power value estimation unit 603, and a computing power allocation unit 604, wherein:
[0126] The computing power concentration unit 601 is used to concentrate computing power resources from different sources into a computing power resource pool, wherein the computing power resources include: physical GPUs.
[0127] The computing power slicing and monitoring unit 602 is used to slice the physical GPU into VGPUs of any size and perform real-time dynamic monitoring of the status of each VGPU.
[0128] The computing power value estimation unit 603 is used to systematically divide the target task when it is necessary to call part of the computing power resources to execute the target task, estimate the computing power value required to complete the target task, and perform range expansion calculations based on the adapted operating carrier to allocate sufficient computing power to the target task.
[0129] The computing power allocation unit 604 is used to directly call a corresponding amount of computing power resources from a computing power resource pool and allocate them to the target task after completing the computing power value estimation.
[0130] Exemplarily, when there are multiple tasks being performed simultaneously, the computing power allocation unit 604 is further used to: perform cross-type resource configuration, use the running start and end time of the computing power resources as the configuration division rule, and perform adaptive configuration of the same area;
[0131] The computing power division and monitoring unit 602 is also used to dynamically monitor the status of computing power resources in real time during the process of computing power resource allocation and use.
[0132] Exemplarily, the above-mentioned device may also include: a computing power recovery unit 605, wherein the computing power recovery unit 605 is used to automatically return the corresponding computing power resources to the computing power resource pool after the target task is completed, so as to wait for secondary configuration.
[0133] Each module in the above computing power scheduling device can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.
[0134] In an exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be shown in FIG7. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit, and an input device. The processor, the memory, and the input / output interface are connected via a system bus, and the communication interface, the display unit, and the input device are connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be implemented through WIFI, a mobile cellular network, near field communication (NFC) or other technologies. When the computer program is executed by the processor, a computing power scheduling method is implemented. The display unit of the computer device is used to form a visually visible picture, which may be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.
[0135] Those skilled in the art will understand that the structure shown in FIG. 7 is merely a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.
[0136] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:
[0137] Concentrate computing resources from different sources into a computing resource pool, wherein the computing resources include: physical GPUs;
[0138] Slice the physical GPU into VGPUs of any size and dynamically monitor the status of each VGPU in real time;
[0139] When part of the computing power resources need to be called to perform the target task, the target task is systematically tapped, the computing power value required to complete the target task is estimated, and the scope is expanded according to the carrier of the adapted operation to allocate sufficient computing power to the target task;
[0140] After the computing power value estimation is completed, the corresponding amount of computing power resources are directly called through the computing power resource pool to be allocated to the target task.
[0141] In one embodiment, when the processor executes the computer program, the following steps are also implemented:
[0142] When multiple target tasks are being carried out simultaneously, cross-resource configuration is performed, with the start and end time of computing resources as the configuration division rule, and adaptive configuration of the same area is performed;
[0143] During the process of computing resource allocation and use, the status of computing resources is monitored dynamically in real time;
[0144] After the target task is completed, the corresponding computing resources are automatically returned to the computing resource pool to wait for secondary configuration.
[0145] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:
[0146] Concentrate computing resources from different sources into a computing resource pool, wherein the computing resources include: physical GPUs;
[0147] Slice the physical GPU into VGPUs of any size and dynamically monitor the status of each VGPU in real time;
[0148] When part of the computing power resources need to be called to perform the target task, the target task is systematically tapped, the computing power value required to complete the target task is estimated, and the scope is expanded according to the carrier of the adapted operation to allocate sufficient computing power to the target task;
[0149] After the computing power value estimation is completed, the corresponding amount of computing power resources are directly called through the computing power resource pool to be allocated to the target task.
[0150] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:
[0151] When multiple target tasks are being carried out simultaneously, cross-resource configuration is performed, with the start and end time of computing resources as the configuration division rule, and adaptive configuration of the same area is performed;
[0152] During the process of computing resource allocation and use, the status of computing resources is monitored dynamically in real time;
[0153] After the target task is completed, the corresponding computing resources are automatically returned to the computing resource pool to wait for secondary configuration.
[0154] In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps:
[0155] Concentrate computing resources from different sources into a computing resource pool, wherein the computing resources include: physical GPUs;
[0156] Slice the physical GPU into VGPUs of any size and dynamically monitor the status of each VGPU in real time;
[0157] When part of the computing power resources need to be called to perform the target task, the target task is systematically tapped, the computing power value required to complete the target task is estimated, and the scope is expanded according to the carrier of the adapted operation to allocate sufficient computing power to the target task;
[0158] After the computing power value estimation is completed, the corresponding amount of computing power resources are directly called through the computing power resource pool to be allocated to the target task.
[0159] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:
[0160] When multiple target tasks are being carried out simultaneously, cross-resource configuration is performed, with the start and end time of computing resources as the configuration division rule, and adaptive configuration of the same area is performed;
[0161] During the process of computing resource allocation and use, the status of computing resources is monitored dynamically in real time;
[0162] After the target task is completed, the corresponding computing resources are automatically returned to the computing resource pool to wait for secondary configuration.
[0163] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.
[0164] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.
[0165] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.
[0166] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.
Claims
1. A computing power scheduling system, characterized in that: The system includes: a resource pool module, a computing power control module, a functional component module, a component communication module, and a deployment form module, wherein: The resource pool module is used to build a computing resource pool, introduce computing resources into the computing resource pool, and classify and centrally manage the introduced computing resources; The computing power control module is used to allocate the computing power resources of the computing power resource pool in a fragmented manner; The functional component module is established in the computing resource pool architecture constructed by the resource pool module, and is used to complete the computing power sharing between different functional components; The component communication module is used to complete the communication between different functional components; The deployment form module is used to integrate multiple container cloud platforms to complete cross-platform computing power scheduling.
2. The system according to claim 1, characterized in that The resource pool module is provided with a plurality of resource nodes, through which computing resources are introduced into the computing resource pool; wherein the computing resources include: a physical graphics processing unit GPU, a virtual graphics processing unit VGPU; The resource pool module is specifically used for: Decoupling an artificial intelligence AI application from a physical graphics processing unit GPU, and providing a virtual graphics processing unit VGPU for the AI application; The computing resources are dynamically scheduled and the status of the computing resources is monitored in real time.
3. The system according to claim 2, characterized in that The computing power control module is specifically used for: The computing resources introduced by the resource pool module are optimized, and the physical GPU is sliced into VGPUs of any size.
4. The system according to claim 1, characterized in that The functional component module includes at least one centralized component control core, and the functional component module is specifically used for: Connecting each functional component to the component control core via a network and maintaining information synchronization between the functional components; Aggregate various information of resource nodes to the component control core, wherein the information includes: any one or more of Internet Protocol IP addresses, physical GPU information, VGPU information, and application task information; The component control core is used to provide any one or more of the following functions: Service registration and service discovery functions of each distributed functional component; Flexible VGPU scheduling and allocation function; Metadata storage and management of multiple copies; License management; Provides various presentation layer state transfer application program interfaces RestAPI required for operation and maintenance.
5. The system according to claim 1, characterized in that The component communication module is specifically used for: The communication between various functional components is established through the management plane network and the data plane network, and the resource pool module is cooperated to complete the management of the computing resource pool and the scheduling of computing resources.
6. The system according to claim 1, characterized in that The deployment form module is specifically used to deploy computing resources on multiple container cloud platforms respectively, wherein the deployment mode includes any of the following: After installing the operating system, deploy each functional component of the computing resource pool directly in binary form; Deploy each functional component of the computing resource pool in a containerized manner.
7. The system according to any one of claims 1 to 6, characterized in that: Also includes: Management plane establishment module, The management plane establishment module is used to establish a management plane through the transmission control protocol / network protocol TCP / IP when deploying the computing power resource pool, so as to complete the visual management of the computing power.
8. A computing power scheduling method, characterized in that: The method is applied to the computing power scheduling system according to any one of claims 1 to 7, and the method comprises: Concentrate computing resources from different sources into a computing resource pool, wherein the computing resources include: physical GPUs; Slice the physical GPU into VGPUs of any size and dynamically monitor the status of each VGPU in real time; When part of the computing power resources need to be called to perform the target task, the target task is systematically tapped, the computing power value required to complete the target task is estimated, and the scope is expanded according to the carrier of the adapted operation to allocate sufficient computing power to the target task; After the computing power value estimation is completed, the corresponding amount of computing power resources are directly called through the computing power resource pool to be allocated to the target task.
9. The method according to claim 8, characterized in that Also includes: When multiple target tasks are being carried out simultaneously, cross-resource configuration is performed, with the start and end time of computing resources as the configuration division rule, and adaptive configuration of the same area is performed; During the process of computing resource allocation and use, the status of computing resources is monitored dynamically in real time; After the target task is completed, the corresponding computing resources are automatically returned to the computing resource pool to wait for secondary configuration.
10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 8 to 9 are implemented.
