Computing power management method of intelligent computing center and related device

Abstract

The invention provides a computing power management method of an intelligent computing center and a related device, and the method comprises the steps: S1, carrying out the standardized conversion of heterogeneous computing resources of the intelligent computing center, and obtaining heterogeneous computing power resources represented in a standardized manner; the heterogeneous computing power resources represented in a standardized manner comprise a resource type, a total capacity and a current usage amount; s2, determining a computing power resource demand of a multi-modal computing task on the intelligent computing center; and S3, allocating the heterogeneous computing power resources to the multi-modal computing task according to a matching corresponding relationship between the resource types and modal data of the multi-modal computing task, the total capacity and the current usage amount of the heterogeneous computing power resources of each resource type, and the computing power resource demand. According to the method, the resource utilization rate of the heterogeneous computing power resources of the intelligent computing center in the multi-modal computing task can be improved.

Description

Technical Field

[0001] Embodiments of the present invention relate to the technical field of computing power infrastructure, and in particular, to a computing power management method and related devices for an intelligent computing center. Background Art

[0002] With the development of artificial intelligence technology and computing power technology, the concept of intelligent computing center has emerged. "Intelligent computing center" refers to the use of large-scale heterogeneous computing power resources, including general computing power and intelligent computing power, mainly for artificial intelligence applications (such as artificial intelligence deep learning model development, model training and model reasoning and other scenarios) to provide the required computing power, data and algorithms. Intelligent computing center covers facilities, hardware, software, and can provide full-stack capabilities from bottom-level computing power to top-level application enablement.

[0003] Intelligent computing centers are facing increasingly complex and diverse computing needs. Intelligent computing centers need to process different types of multimodal data, that is, to perform multimodal computing tasks. The characteristics of these multimodal data are that their computing requirements are different, and they often require the coordinated use of heterogeneous computing resources (for example: CPU (central processing unit), GPU (graphics processing unit), NPU (neural processing unit)) to meet diverse computing performance requirements.

[0004] When processing multimodal data, different tasks may have very different requirements for computing resources. Some tasks require higher computing performance (for example, image processing for deep learning), while others focus more on low-latency responses (for example, real-time video analysis). Existing computing power management methods have low resource utilization when allocating heterogeneous computing power resources in intelligent computing centers to multimodal computing tasks. Summary of the invention

[0005] The embodiment of the present invention provides a computing power management method and related devices for an intelligent computing center to solve the problem of low resource utilization of computing power resources when the existing computing power management method allocates the heterogeneous computing power resources of the intelligent computing center to multimodal computing tasks.

[0006] In order to solve the above-mentioned technical problems, the present invention is achieved as follows:

[0007] In a first aspect, an embodiment of the present invention provides a computing power management method for an intelligent computing center, including:

[0008] Step S1: Standardize and convert the heterogeneous computing resources of the intelligent computing center to obtain standardized heterogeneous computing resources; the standardized heterogeneous computing resources include: resource type, total capacity, and current usage;

[0009] Step S2: Determine the computing resource requirements of the multimodal computing task on the intelligent computing center;

[0010] Step S3: Allocate the heterogeneous computing resources to the multimodal computing task based on the matching correspondence between the resource type and the modal data of the multimodal computing task, the total capacity and the current usage of the heterogeneous computing resources of each resource type, and the computing resource demand.

[0011] Optionally, the step S3 includes:

[0012] Step S31: According to a preset allocation strategy, the allocation of the heterogeneous computing resources to the multimodal computing task is modified in real time;

[0013] The deployment strategy includes at least one of the following strategies: resource matching strategy, load balancing strategy, and priority scheduling strategy.

[0014] Optionally, the deployment strategy includes: the resource matching strategy, the load balancing strategy, and the priority scheduling strategy;

[0015] The step S31 comprises:

[0016] Step S301: multiplying the resource matching strategy, the load balancing strategy, and the priority scheduling strategy by their corresponding weights respectively, and adding the products obtained by the multiplication to obtain a target allocation strategy;

[0017] Step S302: According to the target allocation strategy, the allocation of the heterogeneous computing resources to the multimodal computing task is modified in real time.

[0018] Optionally, the step S301 includes:

[0019] Step S3011: Obtain the real-time computing power resource demand of the multimodal computing task, and adjust the corresponding weights of the resource matching strategy, the load balancing mechanism, and the priority scheduling strategy in real time according to the real-time computing power resource demand.

[0020] Optionally, the expression of the corrected objective function of real-time correction is as follows:

[0021]

[0022] Where T is the set of multimodal computing tasks; CompletionTime(t) is the completion time of multimodal computing task t; IdleTime(r) is the idle time of heterogeneous computing resource r; assigned(t) is the heterogeneous computing resource to which multimodal computing task t is assigned.

[0023] Optionally, the multimodal computing task includes computing tasks on at least one of the following types of modal data: text, image, audio, video, and neural network reasoning.

[0024] In a second aspect, an embodiment of the present invention provides a computing power management device for an intelligent computing center, including:

[0025] A standardization module is used to perform standardized conversion on the heterogeneous computing resources of the intelligent computing center to obtain standardized heterogeneous computing resources; the standardized heterogeneous computing resources include: resource type, total capacity, and current usage;

[0026] A demand determination module, used to determine the computing power resource requirements of the multimodal computing tasks on the intelligent computing center;

[0027] An execution module is used to allocate the heterogeneous computing resources to the multimodal computing task based on the matching correspondence between the resource type and the modal data of the multimodal computing task, the total capacity and the current usage of the heterogeneous computing resources of each resource type, and the computing resource demand.

[0028] In a third aspect, an embodiment of the present invention provides an electronic device comprising a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps in the computing power management method of an intelligent computing center as described in any one of the first aspects.

[0029] In a fourth aspect, an embodiment of the present invention provides a readable storage medium, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps in the computing power management method of an intelligent computing center as described in any one of the first aspects are implemented.

[0030] In a fifth aspect, an embodiment of the present invention provides a computer program product, comprising computer instructions, which, when executed by a processor, implement the steps of the computing power management method of an intelligent computing center as described in any one of the first aspects.

[0031] In an embodiment of the present invention, through step S1: the heterogeneous computing resources of the intelligent computing center are standardized and converted to obtain standardized heterogeneous computing resources; the standardized heterogeneous computing resources include: resource type, total capacity, and current usage; step S2: determining the computing resource requirements of the multimodal computing tasks on the intelligent computing center; step S3: according to the matching correspondence between the resource type and the modal data of the multimodal computing tasks, the total capacity and current usage of the heterogeneous computing resources of each resource type, and the computing resource requirements, the heterogeneous computing resources are deployed to the multimodal computing tasks, which can improve the resource utilization of the heterogeneous computing resources of the intelligent computing center in the multimodal computing tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0033] Figure 1 A flowchart of a computing power management method for an intelligent computing center according to an embodiment of the present invention;

[0034] Figure 2 It is a schematic diagram of the management process of the resource management system;

[0035] Figure 3 This is a principle block diagram of a computing power management device for an intelligent computing center according to an embodiment of the present invention;

[0036] Figure 4 The figure is a principle block diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] The terms "first", "second", etc. of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the terms used in this way are interchangeable where appropriate, so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of one type, and the number of objects is not limited, for example, the first object can be one or more. In addition, "or" in the present application represents at least one of the connected objects. For example, "A or B" covers three schemes, namely, Scheme 1: including A but not including B; Scheme 2: including B but not including A; Scheme 3: including both A and B. The character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0039] In the technical solution of the present disclosure, words such as “connect”, “couple” or “connected” are not limited to physical or mechanical connections, but may include electrical connections.

[0040] In addition, the technical features involved in the different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0041] First, the technical terms involved in the present invention are briefly explained below.

[0042] The "computing power" mentioned in the present invention is the ability of computer equipment or computing / data center to process information. It is the ability of computer hardware and software to work together to execute certain computing requirements. It is the computing power to achieve target result output by processing information data. It is a new type of productivity that integrates information computing power, network carrying capacity, and data storage capacity. It mainly provides services to the society through computing power infrastructure.

[0043] The "computing power" (Computational Power, CP) described in the present invention is the ability of a data center server to process data and output results. It is a comprehensive indicator to measure the computing power of a data center, including general computing power, super computing power and intelligent computing power. The commonly used unit of measurement is the number of floating point operations performed per second (FLOPS: Floating Point Operations Per Second, 1EFLOPS = 10^18FLOPS). The larger the value, the stronger the comprehensive computing power. According to calculations, 1EFLOPS is approximately the computing power output of 5 Tianhe-2A or 500,000 mainstream server CPUs or 2 million mainstream notebooks. The calculation formula is: CP = CP 通用 +CP 智能 +CP 超级 .

[0044] The "Network Power" (NP) described in the present invention is a manifestation of the data transmission capability of computing facilities, including comprehensive capabilities such as network architecture, network bandwidth, transmission latency, intelligent management and scheduling, etc. The carrying capacity involves network transmission within and between data centers, and is a comprehensive indicator for measuring network transmission scheduling capabilities. In the embodiment of the present invention, the carrying capacity uses the memory bandwidth.

[0045] The "Storage Power" (SP) described in the present invention is the comprehensive ability of a data center in terms of data storage capacity, performance, safety and reliability, and green and low-carbon. It is a comprehensive indicator for measuring the data storage capacity of a data center, including external storage devices such as storage arrays and server built-in storage devices. The commonly used unit of measurement for storage capacity is exabyte (EB, 1EB = 2^60bytes), and the commonly used unit of measurement for performance is the number of reads and writes per second per unit capacity (IOPS / TB, Input / Output Operations Per Second / TB). The disaster recovery ratio is an important manifestation of safety and reliability.

[0046] The "computing power infrastructure" described in the present invention is a new type of information infrastructure that integrates information computing power, network carrying capacity, and data storage capacity. It can realize centralized computing, storage, transmission and application of information, and presents characteristics such as diversity and ubiquity, intelligence and agility, security and reliability, and green and low-carbon.

[0047] The “computing power” mentioned in the present invention includes general computing power, intelligent computing power and super computing power.

[0048] The “general computing power” mentioned in the present invention refers to the computing power provided by servers based on CPU (Central Processing Unit) chips, which is used to support basic general computing such as cloud computing and edge computing.

[0049] The "intelligent computing power" described in the present invention is a computing platform for large-scale deployment of special chips such as GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), ASIC (Application Specific Integrated Circuit) for various innovative applications of artificial intelligence, such as natural language processing and machine vision.

[0050] The "super computing power" mentioned in the present invention is mainly the computing power provided by high-performance computing clusters such as supercomputers. It utilizes the centralized computing resources of multiple computer systems working in parallel and uses a dedicated operating system to handle extremely complex or data-intensive problems. It is mainly used for calculations in cutting-edge scientific fields, such as planetary simulation, drug molecule design, and gene analysis.

[0051] The "intelligent computing center" described in the present invention refers to a facility that provides the required computing power, data and algorithms for artificial intelligence applications (such as artificial intelligence deep learning model development, model training and model reasoning scenarios) by using large-scale heterogeneous computing power resources, including general computing power (CPU: Central Processing Unit) and intelligent computing power (GPU: Graphics Processing Unit, FPGA: Field Programmable Gate Array, ASIC: Application Specific Integrated Circuit, etc.). The intelligent computing center covers facilities, hardware, and software, and can provide full-stack capabilities from bottom-level computing power to top-level application enablement.

[0052] The "computing resources" mentioned in the present invention refer to the technologies and facilities with information calculation, transmission, storage and application capabilities required for the development of the digital society, including but not limited to computing resources such as CPU (Central Processing Unit), GPU (Graphics Processing Unit), network resources such as switches and routers, storage resources such as storage arrays and distributed storage, security resources such as firewalls and intrusion detection systems, as well as supporting and guarantee resources such as wind, fire, water and electricity.

[0053] The embodiment of the present invention provides a computing power management method for an intelligent computing center, see Figure 1 As shown, Figure 1 The flowchart of the computing power management method of the intelligent computing center according to the embodiment of the present invention includes:

[0054] Step S1: Standardize and convert the heterogeneous computing resources of the intelligent computing center to obtain standardized heterogeneous computing resources; the standardized heterogeneous computing resources include: resource type, total capacity, and current usage;

[0055] Step S2: Determine the computing resource requirements of the multimodal computing tasks on the intelligent computing center;

[0056] Step S3: Allocate heterogeneous computing resources to multimodal computing tasks based on the matching relationship between resource types and modal data of multimodal computing tasks, the total capacity and current usage of heterogeneous computing resources of each resource type, and computing resource demand.

[0057] It should be noted that heterogeneous computing resources such as CPU, GPU and NPU have their own characteristics and architectural differences. Existing resource management methods are difficult to adapt to the unified management of all types of resources. These resources include both general computing resources (such as CPU) and high-performance computing resources (such as GPU and NPU), and their scheduling requirements are significantly different. These resources are managed independently and difficult to be uniformly scheduled, resulting in the inability of traditional resource management systems to effectively coordinate the use of multiple resources, limiting the overall computing performance of data centers.

[0058] Based on this, the embodiment of the present invention executes step S1: standardizes the heterogeneous computing resources of the intelligent computing center to obtain standardized heterogeneous computing resources; the standardized heterogeneous computing resources include: resource type, total capacity, and current usage. The heterogeneous computing resources are abstracted into a unified resource object (that is, standardized heterogeneous computing resources), which lays the foundation for the unification of resource management. It not only breaks through the management barriers of heterogeneous resources, realizes the collaborative management of multiple types of resources, but also provides standardized support for the dynamic scheduling of resources, which is conducive to improving the utilization rate of heterogeneous computing resources.

[0059] In an embodiment of the present invention, step S2: determining the computing power resource requirements of the multimodal computing tasks on the intelligent computing center, that is, determining the computing power resource requirements of the computing tasks of different modes in the multimodal computing tasks. It should be noted that the distinction between modalities is based on the type of data processed by the computing tasks (that is, modal data). For example, two computing tasks that process data of text and video correspond to two different modalities.

[0060] It should be noted that the computing power resource demand may include the type of computing power resources required and the amount of computing power resources required.

[0061] In step S3 of the embodiment of the present invention, a computing task in which the data type of the processed data is text is taken as an example. The modal data is text, and there is a matching correspondence with the resource type CPU, that is, the CPU needs to be allocated to the computing task in which the modal data is text. According to the total capacity and current usage of the resource type CPU computing power resources, the deployable CPU computing power resources are determined. According to the computing power resource requirements of the computing task in which the modal data is text, the deployable CPU computing power resources are allocated to the computing task in which the modal data is text according to the demand. It should be noted that the above is only exemplary, and the specific execution steps corresponding to the computing tasks of other modal data are similar and can be inferred, and will not be repeated here.

[0062] It can be understood that, roughly speaking, computing tasks for processing data whose data type is text can be completed by CPU resources, and computing tasks for processing data whose data type is video can be completed by GPU resources, thereby constituting different computing resource requirements for multimodal computing tasks.

[0063] However, from a more refined perspective, there is also overlap in the computing resources required for computing tasks of different modalities. For example: the computing resource requirements for computing tasks that process data whose data type is text include CPU and GPU (that is, the required computing resource types are CPU and GPU. The text also includes some pictures, so GPU is actually also required) resources. And the computing resource requirements for computing tasks that process data whose data type is video also include CPU and GPU. Therefore, for computing tasks that process data whose data type is text, only CPU resources cannot be allocated, and for computing tasks that process data whose data type is video, only GPU resources cannot be allocated. Even for computing tasks of single modality data, the resource dimensions that need to be coordinated and allocated have increased. The embodiments of the present invention can ensure that the resources required for computing tasks of each modality can be guaranteed.

[0064] Thus, the embodiment of the present invention can allocate CPU resources to computing tasks whose data type is text, and allocate GPU resources to computing tasks whose video type is text. Specifically, in step S3 of the present embodiment, the computing task whose data type is text is taken as an example. The modal data is text, and there is a matching correspondence relationship with the resource types CPU and GPU, that is, the CPU and GPU need to be allocated to the computing task whose modal data is text. According to the total capacity and current usage of the computing power resources of the resource types CPU and GPU, the deployable CPU and GPU computing power resources are determined. According to the computing power resource requirements of the computing task whose modal data is text, the deployable CPU and GPU computing power resources are deployed to the computing task whose modal data is text according to the demand. It should be noted that the above is only exemplary. For computing tasks of other modal data, for example, the computing power resource requirements of the computing task whose data type is video include CPU and GPU, and the computing power resource requirements of the computing task of neural network reasoning include CPU, GPU and NPU. The specific execution steps are similar and can be inferred by analogy, and will not be repeated here.

[0065] In an embodiment of the present invention, through step S1: the heterogeneous computing resources of the intelligent computing center are standardized and converted to obtain standardized heterogeneous computing resources; the standardized heterogeneous computing resources include: resource type, total capacity, and current usage; step S2: determining the computing resource requirements of the multimodal computing tasks on the intelligent computing center; step S3: according to the matching correspondence between the resource type and the modal data of the multimodal computing tasks, the total capacity and current usage of the heterogeneous computing resources of each resource type, and the computing resource requirements, the heterogeneous computing resources are deployed to the multimodal computing tasks, which can improve the resource utilization of the heterogeneous computing resources of the intelligent computing center in the multimodal computing tasks.

[0066] In some embodiments of the present invention, optionally, step S3 includes:

[0067] Step S31: According to a preset allocation strategy, the allocation of heterogeneous computing resources in multimodal computing tasks is corrected in real time;

[0068] Among them, the deployment strategy includes at least one of the following strategies: resource matching strategy, load balancing strategy, and priority scheduling strategy.

[0069] It should be noted that for intelligent computing centers, the total amount of computing resources is fixed. Therefore, when faced with a large number of multimodal computing tasks, it is often impossible to fully meet the computing resource requirements at the same time (if they are fully met at the same time, the total amount of computing resources is insufficient). Therefore, it is necessary to coordinate and allocate computing resources according to the allocation strategy.

[0070] Resource matching strategy: that is, matching resources based on the computing resource requirements of the computing task (for example, computing type and capacity). The system selects matching resource objects according to the computing type required for the task and sorts them by resource utilization to ensure that resources with low utilization are allocated first, thereby reducing resource idle time and improving resource utilization efficiency.

[0071] Load balancing mechanism: To avoid performance degradation caused by uneven load in a distributed environment, the present invention adopts a load balancing mechanism. When allocating resources, the algorithm adjusts the priority of task allocation according to the utilization of each resource, and prioritizes allocation to resources with lower loads to achieve uniform distribution of resources. The load balancing mechanism ensures the stable operation of distributed tasks and significantly improves the overall resource utilization efficiency of the data center.

[0072] Priority scheduling strategy: For tasks of different priorities (the priority can be preset by the user), by assigning priority weights to tasks, the algorithm can give priority to high-priority tasks when resources are tight, so as to reduce the waiting time of key tasks and improve the response speed of tasks. The priority scheduling mechanism ensures the priority of important tasks in resource allocation and improves the overall service quality of the system.

[0073] In an embodiment of the present invention, through step S31: according to a preset allocation strategy, the allocation of heterogeneous computing resources in multimodal computing tasks is corrected in real time; wherein the allocation strategy includes at least one of the following strategies: resource matching strategy, load balancing strategy, priority scheduling strategy, which can realize the coordinated allocation of computing resources among multimodal computing tasks and improve the utilization efficiency of computing resources.

[0074] In some embodiments of the present invention, optionally, the deployment strategy includes: resource matching strategy, load balancing strategy, priority scheduling strategy;

[0075] Step S31 includes:

[0076] Step S301: multiply the resource matching strategy, load balancing strategy, and priority scheduling strategy by their corresponding weights respectively, and add the products obtained by the multiplication to obtain the target allocation strategy;

[0077] Step S302: According to the target allocation strategy, the allocation of heterogeneous computing resources in multimodal computing tasks is corrected in real time.

[0078] In an embodiment of the present invention, the allocation strategy includes: a resource matching strategy, a load balancing strategy, and a priority scheduling strategy. Based on step S301, each allocation strategy is integrated according to the weight to achieve an organic combination of each strategy. Further, through step S302, the organically integrated target allocation strategy is applied to the allocation of heterogeneous computing resources in multimodal computing tasks for real-time correction, which can further improve the utilization efficiency of multimodal computing resources.

[0079] It should be noted that, in some optional embodiments of the present invention, the weights corresponding to the resource matching strategy, the load balancing strategy, and the priority scheduling strategy may be preset by the user.

[0080] In some embodiments of the present invention, optionally, step S301 includes:

[0081] Step S3011: Obtain the real-time computing power resource requirements of the multimodal computing tasks, and adjust the corresponding weights of the resource matching strategy, load balancing mechanism, and priority scheduling strategy in real time according to the real-time computing power resource requirements.

[0082] In the embodiment of the present invention, when the embodiment of the present invention is executed for the first time, the weights corresponding to the resource matching strategy, the load balancing strategy, and the priority scheduling strategy can be preset by the user. After the first execution of the embodiment of the present invention, the subsequent weights can be dynamically modified through step S3011 to ensure that the resource allocation matches the real-time computing power resource requirements of the multimodal computing task, and ensure that the embodiment of the present invention maintains high-precision computing power resource allocation in the long-term computing process, which can further improve the utilization efficiency of multimodal computing power resources.

[0083] In some embodiments of the present invention, optionally, the expression of the modified objective function of real-time modification is as follows:

[0084]

[0085] Where T is the set of multimodal computing tasks; CompletionTime(t) is the completion time of multimodal computing task t; IdleTime(r) is the idle time of heterogeneous computing resource r; assigned(t) is the heterogeneous computing resource to which multimodal computing task t is assigned.

[0086] The real-time correction objective function of the real-time correction of the embodiment of the invention is used to guide the real-time correction, which can minimize the completion time of the multimodal computing tasks and minimize the idle time of heterogeneous computing resources, thereby ensuring high resource utilization.

[0087] In some embodiments of the present invention, optionally, the multimodal computing task includes a computing task on at least one of the following types of modal data: text, image, audio, video, and neural network reasoning.

[0088] The following is a description with reference to specific embodiments.

[0089] Abstract different types of computing resources into unified resource objects.

[0090] R = {r i ∣r i =(type i ,capacity i ,usage i )}

[0091] Where: R is the standardized representation of heterogeneous computing resources; type i is the resource type (such as CPU, GPU, NPU); capacity i is the total capacity of the resource; usage i The current usage.

[0092] Resource utilization i )calculate:

[0093]

[0094] See also Figure 2 As shown, Figure 2 The diagram is a management process diagram of a resource management system (hereinafter referred to as the management system or the system), and the management process includes:

[0095] 1. Get Started

[0096] The process starts from the "Start" node and enters the first step of the unified resource management method.

[0097] 2. Definition of resource abstract model

[0098] In this step, the system first defines a resource abstraction model. The core idea of ​​this model is to abstract different types of heterogeneous computing resources (such as CPU, GPU, NPU, etc.) into a unified resource object, thereby shielding the differences in resources and enabling subsequent management and scheduling to be unified.

[0099] 3. Check resource type

[0100] After the resource model is defined, the system will check the resource type and classify the resources according to the characteristics of different resource types.

[0101] CPU: If the resource type is CPU, go to the CPU resource abstraction step.

[0102] GPU: If the resource type is GPU, enter the GPU resource abstraction step.

[0103] NPU: If the resource type is NPU, enter the NPU resource abstraction step.

[0104] By checking the resource type, the system can develop a unified abstract standard for different resources.

[0105] 4. Abstract different resource types into unified resource objects

[0106] Based on the inspection results, the system abstracts different computing resource types into standardized resource objects (also known as resource types). The specific steps are as follows:

[0107] CPU resource object: The system abstracts CPU resources into a unified resource object and records its total capacity, current usage and other attributes.

[0108] GPU resource object: The system abstracts GPU resources into a unified resource object and also records related properties.

[0109] NPU resource object: The system abstracts NPU resources into a unified resource object and unifies its attribute description.

[0110] Figure 2 It simplifies the complexity of heterogeneous resource management by abstracting different types of resources and converting them into standardized resource objects.

[0111] 5. Unified description of total resource capacity and current usage

[0112] After resource abstraction, the system will uniformly describe the total capacity and current usage of each resource object. This description method provides standardized quantitative indicators for different types of resources, which facilitates subsequent resource allocation and scheduling.

[0113] Total capacity: The total computing power of a resource, such as the number of cores of a CPU, the computing power (such as TFLOPS) of a GPU, or the processing performance of an NPU.

[0114] Current usage: The resource usage of the current task, indicating the real-time load of the resources.

[0115] Through this unified description, the system can better quantify resource utilization and provide important basic data for scheduling decisions.

[0116] 6. Calculate resource utilization based on resource description formula

[0117] After describing the total resource capacity and current usage, the system will calculate the resource utilization rate based on the resource description formula. Resource utilization rate is an important indicator to measure the current usage status of resources. The calculation formula is:

[0118]

[0119] The calculation results of resource utilization can help the system track resource utilization in real time, ensuring timely adjustments when resource load is too high or too low, thereby optimizing resource allocation efficiency.

[0120] 7. Feedback resource data to the scheduling module

[0121] After calculating the resource utilization, the system feeds back the resource data to the scheduling module. The feedback data includes the current status of each resource (such as utilization and remaining capacity), and the scheduling module performs reasonable task scheduling based on this data.

[0122] Resource status feedback: Resource status data will be passed to the scheduling module, which will select appropriate resources based on the resource load to achieve task load balancing and efficient use of resources.

[0123] 8. Resource management completed

[0124] At this point, the system has completed the unified resource management process, and the abstraction, description, status update and data feedback of resources have all been completed. This process ensures the unified management of multiple heterogeneous computing resources by the system, making the distributed scheduling of resources more efficient and meeting the resource requirements of multimodal data computing.

[0125] Scheduling strategy is a core component in distributed management systems. It aims to improve resource utilization, reduce task waiting time, and ensure priority execution of important tasks through reasonable resource allocation strategies. Scheduling strategy includes the following three key parts: resource matching strategy, load balancing, and priority scheduling.

[0126] A. Resource matching strategy

[0127] Resource matching is the first step in the scheduling strategy, which mainly allocates the most appropriate resource type based on the specific requirements of the task for computing resources. Different tasks may have different requirements for computing resources, for example:

[0128] Image processing tasks usually require high parallel processing capabilities and are suitable for using GPUs;

[0129] Neural network inference tasks may be better suited for NPUs with deep learning acceleration capabilities;

[0130] Text processing (text does not include images) or data preprocessing tasks can be completed using general CPU resources.

[0131] The system first analyzes the task requirements, identifies the resource types required for the task (such as CPU, GPU, or NPU), and then selects qualified computing resources from the resource pool. Through this precise matching, the system can avoid unnecessary occupation of resources while meeting the task requirements, thereby improving the efficiency of resource allocation and fully utilizing computing power.

[0132] B. Load balancing strategy

[0133] After completing the matching of resource types, the scheduling system enters the load balancing stage to ensure that the utilization and efficiency of system resources are optimal. The specific process of load balancing includes the following aspects:

[0134] Utilization sorting: The system will sort the matched resources according to their current utilization, giving priority to resources with lower utilization for task allocation. This method can effectively avoid the situation where some resources are overloaded while other resources are idle, and improve the overall balance of resources.

[0135] Dynamic load adjustment: The system monitors resource utilization in real time and dynamically adjusts the load according to the execution status of the task. For resources with high load, the system will appropriately reduce the task allocation and prioritize the tasks to resources with lower load to avoid excessive resource congestion and decreased computing performance.

[0136] Resource release and reallocation: After the task is completed, the system will release resources in a timely manner and update the load status to ensure that resources can be quickly put into the next task allocation cycle. This dynamic resource release and reallocation mechanism can further optimize load balancing and ensure the continuous and efficient use of resources.

[0137] Through this multi-level load balancing strategy, the system can effectively avoid concentrated use of resources and long-term idleness, improve task response speed, and maximize resource utilization.

[0138] C. Priority Scheduling Strategy

[0139] Priority scheduling is one of the key parts of the scheduling strategy, which mainly allocates resources based on the importance of the task. The priority scheduling strategy is as follows:

[0140] Task prioritization: The system will sort tasks according to their priority when they enter the allocation process. Usually, high-priority tasks are critical tasks and are given priority in resource allocation. For example, real-time data processing, emergency tasks, and important user tasks are usually given higher priority to ensure that these tasks can receive the most timely computing support.

[0141] Priority resource reservation: In order to ensure the rapid response of high-priority tasks, the system will reserve a portion of resources for high-priority tasks when allocating resources. If a high-priority task appears in the system, these reserved resources will be used first to avoid delays in important tasks caused by resource competition.

[0142] Dynamic priority adjustment: When the system load changes, the administrator can dynamically adjust the priority of the task to cope with the changes in real-time task requirements. For example, during system peak hours, important tasks can be assigned first, and low-priority tasks can be appropriately delayed to ensure that the system can still respond to critical tasks under high load.

[0143] Through the priority scheduling strategy, the system can effectively ensure the priority response and completion of key tasks, effectively control the reasonable allocation order of resources, and ensure that tasks are supported by computing resources in order of importance.

[0144] The effects of the management system of the present invention are mainly reflected in several key aspects such as resource utilization, operating costs, computing performance, and management efficiency, providing substantial performance improvement and economic benefits for the operation of the data center.

[0145] 100. Improve resource utilization

[0146] The intelligent dynamic scheduling algorithm of the present invention plays an important role in improving resource utilization. Through accurate matching, real-time monitoring and dynamic allocation of resources, the system can achieve efficient scheduling among various computing resources, avoiding long-term idleness or overload of resources. The specific effects are as follows:

[0147] Dynamic resource adjustment: Real-time monitoring of resource utilization and dynamic adjustment of resource allocation based on task load ensures that the system always maintains the optimal resource configuration state. This minimizes the idle time of low-load resources.

[0148] Balanced resource load: The scheduling algorithm uses a load balancing strategy to reasonably allocate tasks to resources with lower utilization, ensuring balanced use of various computing resources and reducing performance bottlenecks caused by excessive use.

[0149] Reduce resource contention: Through task priority sorting and resource matching strategies, the system reduces resource competition under high load and further improves the overall resource allocation efficiency.

[0150] 200. Reduce operating costs

[0151] The present invention not only improves the computing power of the data center in terms of performance, but also effectively controls the operating costs by optimizing the utilization rate and service life of resources. Specific effects include:

[0152] Delaying the need for hardware expansion: Efficient resource allocation and load balancing reduce the demand for new hardware, avoid frequent hardware expansion due to uneven resource allocation, and save equipment procurement and maintenance costs for data centers.

[0153] Reduce energy consumption: Efficient use of resources enables each device to work under a reasonable load, reducing energy waste from idle devices, thereby significantly reducing the energy consumption of the data center, in line with the concept of green computing.

[0154] Extend the service life of hardware: The system optimizes resource allocation under high load to prevent the hardware from running at full load for a long time, thus extending the normal use cycle of the equipment and reducing the replacement cost caused by hardware loss.

[0155] 300. Improve computing performance

[0156] This invention greatly improves the computing performance of the data center through a reasonable scheduling strategy, ensuring that tasks can be responded to quickly under high concurrency conditions. The specific effects are as follows:

[0157] Reduce task waiting time: Through priority scheduling, the system gives priority to high-priority tasks when resources are tight, reducing the waiting time of critical tasks and meeting the needs of real-time computing.

[0158] Optimizing multimodal data processing: The resource matching and task allocation strategies of the present invention adapt to the requirements of different types of multimodal data tasks. For example, tasks such as images, text, audio, and video are processed in parallel on adapted resources, greatly improving the task processing speed.

[0159] Support for high-concurrency tasks: The system maintains efficient task allocation and scheduling capabilities under high load conditions, supports the execution of a large number of parallel tasks, and effectively improves the processing capacity of the data center during peak periods.

[0160] 400. Enhance management capabilities

[0161] The unified management platform of the present invention provides administrators with convenient system control functions through a visual resource management interface, greatly improving management efficiency. Specifically,

[0162] Real-time monitoring and management: The platform provides real-time monitoring of resource status and task execution. Administrators can view data such as resource utilization and task status of the current system at any time, providing a basis for system adjustment and maintenance.

[0163] Customized scheduling policies: Administrators can flexibly set scheduling policies, such as priority scheduling policies, load balancing policies, etc., according to different business needs to adapt to different task loads and business goals.

[0164] Historical data analysis and optimization: The platform has data analysis capabilities. By analyzing historical tasks and resource utilization, it provides data support for future resource allocation, making system management more scientific and data-driven, and improving long-term resource management efficiency.

[0165] 500. Enhance the scalability and adaptability of the system

[0166] The present invention makes the system more scalable and adaptable through resource abstraction and unified management mode, and can support the future expansion of the data center and changes in computing needs. Specific effects include:

[0167] Compatible with multiple computing resources: The system can adapt to the access and management of multiple heterogeneous computing resources, such as CPU, GPU, NPU, etc., to ensure that new resources can be seamlessly added to the system in the future and realize flexible expansion of resources.

[0168] Adapt to changes in task requirements: The system's scheduling strategy adapts to different task requirements. Whether it is a real-time response task or a batch data processing task, it can be efficiently scheduled in the system, improving the system's versatility and adaptability.

[0169] The embodiment of the present invention provides a computing power management device for an intelligent computing center, see Figure 3 As shown, Figure 3 This is a principle block diagram of a computing power management device of an intelligent computing center according to an embodiment of the present invention. The computing power management device 30 of the intelligent computing center includes:

[0170] The standardization module 31 is used to perform standardized conversion on the heterogeneous computing resources of the intelligent computing center to obtain standardized heterogeneous computing resources; the standardized heterogeneous computing resources include: resource type, total capacity, and current usage;

[0171] A demand determination module 32, used to determine the computing resource requirements of the multimodal computing tasks on the intelligent computing center;

[0172] The execution module 33 is used to allocate the heterogeneous computing resources to the multimodal computing task according to the matching correspondence between the resource type and the modal data of the multimodal computing task, the total capacity and the current usage of the heterogeneous computing resources of each resource type, and the computing resource demand.

[0173] In some embodiments of the present invention, optionally, the execution module 33 is further used to modify the allocation of the heterogeneous computing resources to the multimodal computing task in real time according to a preset allocation strategy;

[0174] The deployment strategy includes at least one of the following strategies: resource matching strategy, load balancing strategy, and priority scheduling strategy.

[0175] In some embodiments of the present invention, optionally, the deployment strategy includes: the resource matching strategy, the load balancing strategy, and the priority scheduling strategy;

[0176] The execution module 33 is further used to multiply the resource matching strategy, the load balancing strategy, and the priority scheduling strategy by their corresponding weights, and add the products obtained by the multiplication to obtain a target allocation strategy;

[0177] The execution module 33 is also used to modify the allocation of the heterogeneous computing resources to the multimodal computing task in real time according to the target allocation strategy.

[0178] In some embodiments of the present invention, optionally, the execution module 33 is also used to obtain the real-time computing power resource demand of the multimodal computing task, and adjust the corresponding weights of the resource matching strategy, the load balancing mechanism, and the priority scheduling strategy in real time according to the real-time computing power resource demand.

[0179] In some embodiments of the present invention, optionally, the expression of the modified objective function of real-time modification is as follows:

[0180]

[0181] Where T is the set of multimodal computing tasks; CompletionTime(t) is the completion time of multimodal computing task t; IdleTime(r) is the idle time of heterogeneous computing resource r; assigned(t) is the heterogeneous computing resource to which multimodal computing task t is assigned.

[0182] In some embodiments of the present invention, optionally, the multimodal computing task includes a computing task for at least one of the following types of modal data: text, image, audio, video, and neural network reasoning.

[0183] The computing power management device of the intelligent computing center provided in the embodiment of the present application can achieve Figure 1 to Figure 2 The various processes implemented by the method embodiment and achieving the same technical effect are not described here to avoid repetition.

[0184] An embodiment of the present invention provides an electronic device 40, see Figure 4 As shown, Figure 4 This is a principle block diagram of an electronic device 40 according to an embodiment of the present invention, including a processor 41, a memory 42, and a program or instruction stored in the memory 42 and executable on the processor 41. When the program or instruction is executed by the processor, the steps in the computing power management method of any intelligent computing center of the present invention are implemented.

[0185] An embodiment of the present invention provides a readable storage medium, on which programs or instructions are stored. When the programs or instructions are executed by a processor, the various processes of an embodiment of a computing power management method for an intelligent computing center such as any of the above-mentioned items are implemented, and the same technical effect can be achieved. To avoid repetition, they will not be repeated here.

[0186] The readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc. In some examples, the readable storage medium may be a non-transitory readable storage medium.

[0187] An embodiment of the present invention also provides a computer program product, including computer instructions. When the computer instructions are executed by a processor, the various processes of any of the above-mentioned embodiments of the computing power management method of the intelligent computing center can be implemented, and the same technical effect can be achieved. To avoid repetition, they will not be repeated here.

[0188] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

Claims

1. A computing power management method for an intelligent computing center, characterized in that: include: Step S1: Standardize and convert the heterogeneous computing resources of the intelligent computing center to obtain standardized heterogeneous computing resources; The standardized representation of heterogeneous computing resources includes: resource type, total capacity, and current usage; Step S2: Determine the computing resource requirements of the multimodal computing task on the intelligent computing center; Step S3: Allocate the heterogeneous computing resources to the multimodal computing task based on the matching correspondence between the resource type and the modal data of the multimodal computing task, the total capacity and the current usage of the heterogeneous computing resources of each resource type, and the computing resource demand.

2. The computing power management method of the intelligent computing center according to claim 1, characterized in that: The step S3 comprises: Step S31: According to a preset allocation strategy, the allocation of the heterogeneous computing resources to the multimodal computing task is modified in real time; The deployment strategy includes at least one of the following strategies: resource matching strategy, load balancing strategy, and priority scheduling strategy.

3. The computing power management method of the intelligent computing center according to claim 2, characterized in that: The deployment strategy includes: the resource matching strategy, the load balancing strategy, and the priority scheduling strategy; The step S31 comprises: Step S301: multiplying the resource matching strategy, the load balancing strategy, and the priority scheduling strategy by their corresponding weights respectively, and adding the products obtained by the multiplication to obtain a target allocation strategy; Step S302: According to the target allocation strategy, the allocation of the heterogeneous computing resources to the multimodal computing task is modified in real time.

4. The computing power management method of the intelligent computing center according to claim 3 is characterized in that , The step S301 includes: Step S3011: Obtain the real-time computing power resource demand of the multimodal computing task, and adjust the corresponding weights of the resource matching strategy, the load balancing mechanism, and the priority scheduling strategy in real time according to the real-time computing power resource demand.

5. The computing power management method of the intelligent computing center according to claim 2 or 3, characterized in that: The expression of the corrected objective function of real-time correction is as follows: Said Where T is the set of multimodal computing tasks; CompletionTime(t) is the completion time of multimodal computing task t; IdleTime(r) is the idle time of heterogeneous computing resource r; assigned(t) is the heterogeneous computing resource to which multimodal computing task t is assigned.

6. The computing power management method of the intelligent computing center according to claim 1 is characterized in that , The multimodal computing tasks include computing tasks for at least one of the following types of modal data: text, image, audio, video, and neural network reasoning.

7. A computing power management device for an intelligent computing center, characterized in that: include: The standardization module is used to perform standardized conversion on the heterogeneous computing resources of the intelligent computing center to obtain standardized heterogeneous computing resources; The standardized representation of heterogeneous computing resources includes: resource type, total capacity, and current usage; A demand determination module, used to determine the computing power resource requirements of the multimodal computing tasks on the intelligent computing center; An execution module is used to allocate the heterogeneous computing resources to the multimodal computing task based on the matching correspondence between the resource type and the modal data of the multimodal computing task, the total capacity and the current usage of the heterogeneous computing resources of each resource type, and the computing resource demand.

8. An electronic device, characterized in that: It includes a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps in the computing power management method of the intelligent computing center as described in any one of claims 1 to 6.

9. A readable storage medium, characterized in that: The readable storage medium stores programs or instructions, and when the programs or instructions are executed by the processor, the steps in the computing power management method of the intelligent computing center as described in any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that It comprises computer instructions, which, when executed by a processor, implement the steps of the computing power management method of the intelligent computing center as described in any one of claims 1 to 6.

Images