Heterogeneous Computing Power Intelligent Scheduling System and Method
By using a heterogeneous computing power intelligent scheduling system, combined with Kubernetes Operator and CRD, unified management and scheduling of heterogeneous computing power resources are achieved. This solves the compatibility and scalability issues of existing scheduling schemes, meets the business layer's requirements for second-level response and low latency, and improves the utilization efficiency of computing power resources.
Patent Information
- Application Number
- CN202311666674.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-12-06
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2043-12-06
AI Technical Summary
Existing heterogeneous computing power scheduling solutions lack compatibility and scalability, and cannot effectively manage and schedule heterogeneous computing power resources on a unified platform, resulting in the inability to meet the business layer's requirements for second-level response and low latency.
A heterogeneous computing power intelligent scheduling system is adopted, including a computing power scheduling module, a cloud-native storage ETCD module, and a third-party extension strategy module. It provides a unified model and interface library through heterogeneous computing power resources CRD, and combines Kubernetes Operator and CRD to achieve compatibility and extension of scheduling strategies.
It enables unified management and scheduling of heterogeneous computing resources, improves scheduling compatibility and scalability, meets the requirements of business layer for second-level response and low latency, and improves the utilization efficiency of computing resources.
Smart Images

Figure CN117632505B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of computing power task scheduling technology, and in particular to a heterogeneous computing power intelligent scheduling system and method. Background Technology
[0002] As digital transformation deepens across industries, massive amounts of raw data are generated by various industry terminals, necessitating substantial computing power for processing. To meet the requirements of second-level response, low latency, and business continuity at the business layer, computing power clusters are commonly used to provide a highly available, high-throughput, and highly scalable technical architecture. During the annual construction of computing power clusters, numerous heterogeneous computing power cluster resources exist within the same or different data centers. How to manage and schedule these computing power resources on a unified platform, enabling users to conveniently access resources according to their needs and usage habits, is a pressing issue. Currently, mainstream technical solutions for intelligent scheduling of heterogeneous computing power customize a single intelligent scheduling method for each specific scenario. However, each scheduling scheme cannot match the needs of heterogeneous computing power scheduling in other scenarios, and existing schemes lack scalability. Therefore, existing scheduling solutions lack compatibility and scalability. Summary of the Invention
[0003] The main objective of this invention is to provide a heterogeneous computing power intelligent scheduling system and method, which aims to solve the technical problems of lack of compatibility and scalability in existing scheduling schemes.
[0004] To achieve the above objectives, this invention provides a heterogeneous computing power intelligent scheduling system. The system includes a computing power scheduling module, a cloud-native storage ETCD module, and a third-party extension strategy module. The cloud-native storage ETCD module stores heterogeneous computing power resource CRDs, scheduling policy CRDs, computing power task CRDs, and cluster CRDs. The unified model and interface library provided by the heterogeneous computing power resource CRDs enable the embedding of the scheduling policy CRDs within the heterogeneous computing power intelligent scheduling system.
[0005] The third-party extension strategy module is used to register the third-party extension strategy to the cloud-native storage ETCD module through the computing power task interface provided by the heterogeneous computing power resource CRD.
[0006] The cloud-native storage ETCD module is used to update the scheduling policy CRD according to the third-party extension policy to obtain the updated scheduling policy CRD;
[0007] The computing power scheduling module is used to determine the target cluster in the cluster CRD according to the updated scheduling policy CRD, and bind the computing power tasks to the target cluster.
[0008] Optionally, the heterogeneous computing resource CRD is defined based on cloud-native technology. The heterogeneous computing resource CRD includes an interface abstraction layer, which uniformly handles the unit adaptation of computing resources.
[0009] Optionally, the interface abstraction layer includes a Deep Copy interface, a Canonicalize interface for radix normalization, an ADD interface for addition, a SUB interface for subtraction, a CMP interface for comparison, and an Equal interface for equality judgment.
[0010] Optionally, the heterogeneous computing resources CRD are provided with CRUD interfaces through the Kubernetes API server.
[0011] Optionally, the updated scheduling policy CRD includes a built-in policy and a third-party extended policy. The built-in policy is loaded into the cache during the operation of the computing power scheduling module, and the third-party extended policy is read from the cloud-native storage ETCD module by dynamic loading during the operation of the computing power scheduling module.
[0012] Furthermore, to achieve the above objectives, this invention also proposes a heterogeneous computing power intelligent scheduling method, which includes:
[0013] Third-party extension strategies are registered to the cloud-native storage ETCD module through the computing task interface provided by heterogeneous computing resources CRD.
[0014] Update the scheduling policy CRD in the cloud-native storage ETCD module according to the third-party extension policy to obtain the updated scheduling policy CRD;
[0015] The target cluster in the cluster CRD is determined according to the updated scheduling policy CRD, and the computing power tasks are bound to the target cluster.
[0016] Optionally, the updated scheduling policy (CRD) includes built-in policies and third-party extended policies; wherein, determining the target cluster in the cluster CRD according to the updated scheduling policy (CRD) and binding the computing power tasks to the target cluster includes:
[0017] After creating a computing task based on the computing task CRD in the cloud-native storage ETCD module, the computing scheduling module is triggered to start the scheduling workflow, which includes a pre-selection stage, an optimization stage, and a binding stage.
[0018] In the pre-selection phase, the built-in policies are traversed, and the first cluster that satisfies all the built-in policies is selected from the cluster CRD.
[0019] In the selection phase, a second cluster is selected from the first cluster according to the third-party extension strategy;
[0020] During the binding phase, the target cluster in the second cluster is determined, and the computing power task is bound to the target cluster.
[0021] Optionally, the step of selecting the second cluster from the first cluster according to the third-party extension strategy includes:
[0022] S301. Call the third-party Operator / schedule HTTPS interface to execute the target policy in the third-party extended policy, and score the clusters in the first cluster that meet the target policy to obtain the scheduling result;
[0023] S302. Determine whether there is a new target strategy in the third-party extension strategy;
[0024] S303. If yes, then the new target strategy is used as the target strategy in step S301 to repeat steps S301-S303; if no, then the second cluster is selected from the first cluster based on the scheduling result.
[0025] Optionally, selecting the second cluster from the first cluster based on the scheduling result includes:
[0026] Set a preset threshold;
[0027] The weight score of each cluster in the first cluster is determined based on the scheduling results;
[0028] The clusters in the first cluster whose weight scores are greater than the preset threshold are designated as the second cluster.
[0029] Optionally, determining the target cluster in the second cluster and binding the computing power task to the target cluster includes:
[0030] The weight score of each cluster in the second cluster is determined based on the scheduling results;
[0031] The cluster with the highest weight score in the second cluster is selected as the target cluster, and the computing power task is bound to the target cluster.
[0032] This invention proposes a heterogeneous computing power intelligent scheduling system and method. The heterogeneous computing power intelligent scheduling system includes a computing power scheduling module, a cloud-native storage ETCD module, and a third-party extension strategy module. The cloud-native storage ETCD module stores heterogeneous computing power resource CRDs, scheduling policy CRDs, computing power task CRDs, and cluster CRDs. The unified model and interface library provided by the heterogeneous computing power resource CRDs enable the embedded implementation of the scheduling policy CRDs within the heterogeneous computing power intelligent scheduling system. The third-party extension strategy module is used to register third-party extension strategies with the cloud-native storage ETCD module through the computing power task interface provided by the heterogeneous computing power resource CRDs. The cloud-native storage ETCD module is used to update the scheduling policy CRDs according to the third-party extension strategies, obtaining an updated scheduling policy CRD. The computing power scheduling module is used to determine the target cluster in the cluster CRDs based on the updated scheduling policy CRDs and bind computing power tasks to the target clusters. Through the above system, using a unified framework for heterogeneous computing power intelligent scheduling, combined with Kubernetes Operators, CRDs, and built-in strategies, the compatibility and extension functions of heterogeneous computing power scheduling strategies are achieved. Attached Figure Description
[0033] Figure 1 This is a structural block diagram of the first embodiment of the heterogeneous computing power intelligent scheduling system of the present invention;
[0034] Figure 2 This is a schematic diagram illustrating the principle of heterogeneous computing power scheduling in the first embodiment of the heterogeneous computing power intelligent scheduling system of the present invention.
[0035] Figure 3 This is a flowchart illustrating the first embodiment of the heterogeneous computing power intelligent scheduling method of the present invention;
[0036] Figure 4 This is a schematic diagram of the process for determining the second cluster in the first embodiment of the heterogeneous computing power intelligent scheduling method of the present invention;
[0037] Figure 5 This is a scheduling flowchart in the first embodiment of the heterogeneous computing power intelligent scheduling method of the present invention.
[0038] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation
[0039] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.
[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0041] It should be noted that ETCD stands for Distributed Key-Value Store, CRD (Custom Resource Definition) is a mechanism in Kubernetes used to extend the Kubernetes API and allow users to define custom resources, and Operator Controller is a custom controller used to manage and automate specific types of applications or resources in a Kubernetes cluster.
[0042] This invention proposes a heterogeneous computing power intelligent scheduling system.
[0043] Reference Figure 1 In this embodiment of the invention, the heterogeneous computing power intelligent scheduling system includes a computing power scheduling module 10, a cloud-native storage ETCD module 20, and a third-party extension strategy module 30. The cloud-native storage ETCD module 20 stores heterogeneous computing power resource CRDs, scheduling strategy CRDs, computing power task CRDs, and cluster CRDs. The unified model and interface library provided by the heterogeneous computing power resource CRDs enable the embedded implementation of the scheduling strategy CRDs in the heterogeneous computing power intelligent scheduling system. The third-party extension strategy module 30 is used to register third-party extension strategies to the cloud-native storage ETCD module 20 through the computing power task interface provided by the heterogeneous computing power resource CRDs. The cloud-native storage ETCD module 20 is used to update the scheduling strategy CRDs according to the third-party extension strategies to obtain updated scheduling strategy CRDs. The computing power scheduling module 10 is used to determine the target cluster in the cluster CRDs according to the updated scheduling strategy CRDs and bind the computing power tasks to the target clusters.
[0044] It should be noted that scheduling policies include built-in policies and third-party extended policies. Scheduling policies are stored in the cloud-native storage ETCD module using CRD data mode; updated scheduling policy CRD refers to the updated scheduling policy CRD; computing power tasks are stored in the cloud-native storage ETCD module using CRD data mode; multi-cluster information is stored in the cloud-native storage ETCD module using CRD data mode; the computing power scheduling module refers to the computing power scheduling Operator Controller; the third-party extended policy module refers to the third-party extended policy Operator Controller. The third-party extended policy Operator Controller needs to implement the / schedule HTTP interface. The third-party extended policy module is mainly used to register third-party extended policies to the scheduling policy CRD of the cloud-native storage ETCD module.
[0045] Understandably, the heterogeneous computing power intelligent scheduling system can register and trigger dynamic loads at any time during operation; the target cluster can be the cluster with the highest compliance in the cluster CRD, and can be the cluster that needs to be run for this computing power task.
[0046] It should be noted that updating the scheduling policy CRD according to the third-party extended policy is a real-time dynamic update, which can be implemented based on an asynchronous watch mechanism. Specifically, when a third-party user updates the scheduling policy CRD based on the third-party extended policy module, it can be achieved through the following two steps: 1. Based on the unified model and interface library provided by the heterogeneous computing power resource CRD, the third party implements an extended computing power scheduling Operator Controller and provides a standard HTTPS computing power task scheduling interface ( / schedule). HTTPS can ensure communication and data security; 2. Register the third-party extended policy to the cloud-native storage ETCD, triggering the policy cache update of the computing power scheduling Operator Controller. Not only is the extended policy registered, but also the callback access address is registered so that the computing power scheduling Operator Controller knows the address of the / schedule interface call.
[0047] In one embodiment, the heterogeneous computing resource CRD is defined based on cloud-native technology. The heterogeneous computing resource CRD includes an interface abstraction layer, which uniformly handles the unit adaptation of computing resources.
[0048] It should be noted that the interface abstraction layer uniformly handles the unit adaptation of computing resources such as binary, decimal, and floating-point numbers; the interface abstraction layer includes the adaptation of all known computing resource units.
[0049] In this embodiment, the support for different number systems and units ensures compatibility with all heterogeneous computing resources, thus solving the problem of a unified model and measurement for heterogeneous computing power.
[0050] In one embodiment, the interface abstraction layer includes a Deep Copy interface, a Canonicalize interface, an ADD interface, a SUB interface, a CMP interface for comparing sizes, and an Equal interface for determining equality.
[0051] In this embodiment, the Deep Copy interface, Canonicalize interface, ADD interface, SUB interface, CMP interface, and Equal interface in the interface abstraction layer can solve problems such as heterogeneous computing resource conversion, addition, subtraction, and comparison.
[0052] In one embodiment, the heterogeneous computing power resource CRD is provided with CRUD interfaces through a unified k8s apiserver.
[0053] In one embodiment, the updated scheduling policy CRD includes a built-in policy and a third-party extended policy. The built-in policy is loaded into the cache during the operation of the computing power scheduling module, and the third-party extended policy is read from the cloud-native storage ETCD module by dynamic loading during the operation of the computing power scheduling module.
[0054] In specific implementations, such as Figure 2 The diagram showing the heterogeneous computing power scheduling principle indicates that the heterogeneous computing power intelligent scheduling system includes a computing power scheduling Operator Controller, cloud-native storage ETCD, and a third-party extension strategy Operator Controller. Among them, the cloud-native storage ETCD stores heterogeneous computing power resource CRDs, scheduling strategy CRDs, computing power task CRDs, and cluster CRDs. The unified interface abstraction layer of the heterogeneous computing power resource CRDs includes Deep Copy interface, Canonicalize base normalization conversion interface, ADD addition operation interface, SUB subtraction operation interface, CMP size comparison operation interface, and Equal equality judgment operation interface.
[0055] It's important to note that the reference model for heterogeneous computing resource CRDs is a YAML configuration file for a CustomResourceDefinition object in Kubernetes. This object defines a custom resource named "kuberesources.resource.kubearena.io"; its API version is "apiextensions.k8s.io / v1", belonging to the "resource.kubearena.io" group. Its object type is "KubeResource", with the plural form "KubeResourceList", and a short name "res". This custom resource is accessible throughout the cluster and provides a version named "v1"; the definition of this version contains specification and state information about the "KubeResource" object.
[0056] This embodiment utilizes a heterogeneous computing power intelligent scheduling system comprising a computing power scheduling module, a cloud-native storage ETCD module, and a third-party extension strategy module. The cloud-native storage ETCD module stores heterogeneous computing power resource CRDs, scheduling policy CRDs, computing power task CRDs, and cluster CRDs. The unified model and interface library provided by the heterogeneous computing power resource CRDs enable the embedding of the scheduling policy CRDs within the heterogeneous computing power intelligent scheduling system. The third-party extension strategy module registers third-party extension strategies with the cloud-native storage ETCD module through the computing power task interface provided by the heterogeneous computing power resource CRDs. The cloud-native storage ETCD module updates the scheduling policy CRDs according to the third-party extension strategies, obtaining an updated scheduling policy CRD. The computing power scheduling module determines the target cluster in the cluster CRDs based on the updated scheduling policy CRDs and binds computing power tasks to the target clusters. Through this system, using a unified framework for heterogeneous computing power intelligent scheduling, combined with Kubernetes Operators, CRDs, and built-in strategies, the system achieves compatibility and extensibility of heterogeneous computing power scheduling strategies.
[0057] In addition, such as Figure 3 As shown in the figure, this embodiment of the invention also provides a method for intelligent scheduling of heterogeneous computing power.
[0058] In this embodiment, the heterogeneous computing power intelligent scheduling method includes the following steps:
[0059] Step S10: Register the third-party extension strategy to the cloud-native storage ETCD module through the computing task interface provided by the heterogeneous computing resource CRD.
[0060] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a mobile phone, tablet computer, or personal computer, or an electronic device or heterogeneous computing power intelligent scheduling system capable of performing the above functions. The following description uses the heterogeneous computing power intelligent scheduling system as an example to illustrate this embodiment and the subsequent embodiments.
[0061] It should be noted that the computing task interface provided by the heterogeneous computing resource CRD refers to the standard HTTPS computing task scheduling interface ( / schedule);
[0062] Step S20: Update the scheduling policy CRD in the cloud-native storage ETCD module according to the third-party extension policy to obtain the updated scheduling policy CRD.
[0063] Understandably, the heterogeneous computing power intelligent scheduling system can register and trigger dynamic loads at any time during operation; the target cluster can be the cluster with the highest compliance in the cluster CRD, and can be the cluster that needs to be run for this computing power task.
[0064] It should be noted that updating the scheduling policy CRD refers to the updated scheduling policy CRD; updating the scheduling policy CRD includes built-in policies and third-party extended policies; updating the scheduling policy CRD according to the third-party extended policies is a real-time dynamic update, which can be implemented based on an asynchronous watch mechanism. Specifically, when a third-party user updates the scheduling policy CRD based on the third-party extended policy module, it can be achieved through the following two steps: 1. Based on the unified model and interface library provided by the heterogeneous computing power resource CRD, the third party implements an extended computing power scheduling Operator Controller and provides a standard HTTPS computing power task scheduling interface ( / schedule). HTTPS can ensure communication and data security; 2. Register the third-party extended policy to the cloud-native storage ETCD, triggering the policy cache update of the computing power scheduling Operator Controller. Not only the extended policy is registered, but also the callback access address is registered so that the computing power scheduling Operator Controller knows the address of the / schedule interface call.
[0065] Step S30: Determine the target cluster in the cluster CRD according to the updated scheduling policy CRD, and bind the computing power task to the target cluster.
[0066] It should be noted that after binding the computing power task to the target cluster, the current computing power task scheduling is completed. You can continue to create new computing power tasks based on the computing power task CRD stored in the cloud-native storage ETCD module and then schedule them.
[0067] In one embodiment, the Updated Scheduling Policy (CRD) includes a built-in policy and a third-party extended policy; wherein, determining the target cluster in the Cluster CRD according to the Updated Scheduling Policy (CRD) and binding the computing power tasks to the target cluster includes:
[0068] After creating a computing task based on the computing task CRD in the cloud-native storage ETCD module, the computing scheduling module is triggered to start the scheduling workflow, which includes a pre-selection stage, an optimization stage, and a binding stage.
[0069] In the pre-selection phase, the built-in policies are traversed, and the first cluster that satisfies all the built-in policies is selected from the cluster CRD.
[0070] In the selection phase, a second cluster is selected from the first cluster according to the third-party extension strategy;
[0071] During the binding phase, the target cluster in the second cluster is determined, and the computing power task is bound to the target cluster.
[0072] It should be noted that the scheduling workflow consists of three stages: the pre-selection stage, the priority stage, and the binding stage. In the pre-selection stage, all built-in strategies need to be executed. For a single computing task, the built-in strategies are executed sequentially according to the configured order (executing built-in strategies can be understood as determining whether the cluster meets the built-in strategies). The first cluster refers to the cluster in the cluster CRD that meets all built-in strategies. In the priority stage, third-party extension strategies need to be executed (which can be understood as determining whether the cluster meets the third-party extension strategy). The second cluster refers to the cluster in the first cluster that meets the third extension strategy. In the binding stage, the cluster with the highest compliance in the second cluster can be selected as the target cluster, which is the cluster that needs to be run in this computing task.
[0073] It should be noted that when executing a third-party extension strategy, the third-party Operator / scheduleHTTPS interface needs to be called and the scheduling result needs to be received. The parameters of the / schedule interface are the computing task and the previously selected cluster list (cluster CRD), and the returned list is the selected cluster list (i.e., the second cluster).
[0074] In this embodiment, the target cluster is determined from the cluster CRD by scheduling the three stages of the workflow. This not only enables the cluster with the highest compliance to be quickly identified, but also effectively improves the accuracy of scheduling.
[0075] In one embodiment, such as Figure 4As shown, the step of selecting the second cluster from the first cluster according to the third-party extension strategy includes:
[0076] Step S301: Call the third-party Operator / schedule HTTPS interface to execute the target policy in the third-party extended policy, and score the clusters in the first cluster that meet the target policy to obtain the scheduling result.
[0077] Step S302: Determine whether there is a new target strategy in the third-party extension strategy.
[0078] Step S303: If yes, then repeat steps S301-S303 using the new target policy as the target policy in step S301; if no, then select the second cluster from the first cluster based on the scheduling result.
[0079] It is understandable that the new target strategy can be understood as the next third-party extension strategy of the target strategy in the third-party extension strategy.
[0080] It should be noted that the target strategy can be determined based on the registration order of the third-party extension strategies. For example, in the first round of cluster scoring, the target strategy is the first registered strategy among the third-party extension strategies. The scheduling results need to be updated after each round of cluster scoring. For example, in the first round of cluster scoring, since cluster A does not meet the target strategy in the first round of scoring, while cluster B does, the scheduling result obtained after the first round of scoring is: cluster A gets 0 points, and cluster B gets 1 point. In the second round of cluster scoring, since both cluster A and cluster B meet the target strategy in the second round of scoring, the updated scheduling result obtained after the second round of scoring is: cluster A gets 1 point, and cluster B gets 2 points.
[0081] It is understandable that the scheduling result obtained when there is no new target strategy in the third-party extension strategy is the final scheduling result.
[0082] In one embodiment, selecting a second cluster from the first cluster based on the scheduling result includes:
[0083] Set a preset threshold;
[0084] The weight score of each cluster in the first cluster is determined based on the scheduling results;
[0085] The clusters in the first cluster whose weight scores are greater than the preset threshold are designated as the second cluster.
[0086] It should be noted that the scheduling result refers to the final scheduling result, which can be used to determine the weight score of each cluster in the first cluster. The preset threshold can be set in advance. Preferably, the preset threshold can be set to 0. Since the weight score of a cluster is greater than or equal to 0, when the weight score of a cluster is not 0, the cluster can be regarded as the second cluster. The cluster with a weight score of 0 can be regarded as a cluster that does not meet any of the target strategies in the third-party extension strategy.
[0087] In this embodiment, the second cluster is determined based on the scoring of each cluster in the first cluster by setting a threshold, which can effectively improve the determination efficiency of the second cluster.
[0088] In one embodiment, determining the target cluster in the second cluster and binding the computing power task to the target cluster includes:
[0089] The weight score of each cluster in the second cluster is determined based on the scheduling results;
[0090] The cluster with the highest weight score in the second cluster is selected as the target cluster, and the computing power task is bound to the target cluster.
[0091] It should be noted that the scheduling result refers to the final scheduling result, which can be used to determine the weight score of each cluster in the second cluster. Specifically, for example, if the weight score of each cluster in the second cluster is determined based on the scheduling result: the weight score of cluster A is 1 point and the weight score of cluster B is 2 points, then cluster B is the target cluster in the second cluster.
[0092] It should be noted that after binding the computing power task to the target cluster, the current computing power task scheduling is completed. You can continue to create new computing power tasks based on the computing power task CRD stored in the cloud-native storage ETCD module and then schedule them.
[0093] In this embodiment, by selecting the cluster with the highest weight score as the cluster to be run in this computing task, the accuracy of computing power scheduling can be effectively improved.
[0094] In specific implementations, such as Figure 5 As shown, it is necessary to define a heterogeneous computing resource CRD based on cloud-native technology, and implement a unified interface abstraction layer through the heterogeneous computing resource CRD to complete the unified model and measurement framework of heterogeneous computing power. Then, based on the unified model and interface library provided by the heterogeneous computing resource CRD, a built-in scheduling strategy can be implemented. Then, a computing task can be created to trigger the scheduling workflow. The scheduling of this computing task can be completed based on the three stages of the scheduling workflow.
[0095] It should be understood that the above are merely illustrative examples and do not constitute any limitation on the technical solutions of the present invention. In specific applications, those skilled in the art can make settings as needed, and the present invention does not impose any restrictions on this.
[0096] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of this invention. In practical applications, those skilled in the art can select some or all of the workflow to achieve the purpose of this embodiment according to actual needs, and no restrictions are imposed here.
[0097] In addition, for technical details not described in detail in this embodiment, please refer to the heterogeneous computing power intelligent scheduling method provided in any embodiment of the present invention, which will not be repeated here.
[0098] Furthermore, it should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.
[0099] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.
[0100] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory (ROM) / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.
[0101] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.
Claims
1. A heterogeneous computing power intelligent scheduling method, characterized in that, include: Third-party extension strategies are registered to the cloud-native storage ETCD module through the computing task interface provided by heterogeneous computing resources CRD. The scheduling policy CRD in the cloud-native storage ETCD module is updated according to the third-party extended policy to obtain the updated scheduling policy CRD, which includes built-in policies and third-party extended policies. The target cluster in the cluster CRD is determined according to the updated scheduling policy CRD, and the computing power tasks are bound to the target cluster. After a computing task is created based on the computing task CRD in the cloud-native storage ETCD module, the computing scheduling module is triggered to start the scheduling workflow, which includes a pre-selection stage, an optimization stage, and a binding stage. In the pre-selection stage, the built-in strategies are traversed to select a first cluster that satisfies all built-in strategies from the cluster CRD. In the optimization stage, a second cluster is selected from the first cluster based on the third-party extension strategy. In the binding stage, a target cluster in the second cluster is determined, and the computing task is bound to the target cluster. The step of selecting a second cluster from the first cluster based on the third-party extension strategy includes: S301, calling the third-party Operator / schedule HTTPS interface to execute the target strategy in the third-party extension strategy, and scoring the clusters in the first cluster that meet the target strategy to obtain a scheduling result; S302, determining whether there is a new target strategy in the third-party extension strategy; S303, if yes, repeating steps S301-S303 as the target strategy in step S301; if no, selecting a second cluster from the first cluster based on the scheduling result, wherein a preset threshold is set; the weight score of each cluster in the first cluster is determined according to the scheduling result; and the clusters in the first cluster whose weight score is greater than the preset threshold are selected as the second cluster.
2. The method of claim 1, wherein, The step of determining the target cluster in the second cluster and binding the computing power task to the target cluster includes: The weight score of each cluster in the second cluster is determined based on the scheduling results; The cluster with the highest weight score in the second cluster is selected as the target cluster, and the computing power task is bound to the target cluster.
3. A heterogeneous computing power intelligent scheduling system for implementing the heterogeneous computing power intelligent scheduling method of any one of claims 1-2, characterized in that, The heterogeneous computing power intelligent scheduling system includes a computing power scheduling module, a cloud-native storage ETCD module, and a third-party extension strategy module. The cloud-native storage ETCD module stores heterogeneous computing power resource CRDs, scheduling policy CRDs, computing power task CRDs, and cluster CRDs. The unified model and interface library provided by the heterogeneous computing power resource CRDs enable the embedding of the scheduling policy CRDs within the heterogeneous computing power intelligent scheduling system. The third-party extension strategy module is used to register the third-party extension strategy to the cloud-native storage ETCD module through the computing power task interface provided by the heterogeneous computing power resource CRD. The cloud-native storage ETCD module is used to update the scheduling policy CRD according to the third-party extension policy to obtain the updated scheduling policy CRD; The computing power scheduling module is used to determine the target cluster in the cluster CRD according to the updated scheduling policy CRD, and bind the computing power tasks to the target cluster.
4. The system of claim 3, wherein, The heterogeneous computing resource CRD is defined based on cloud-native technology. The heterogeneous computing resource CRD includes an interface abstraction layer, which uniformly handles the unit adaptation of computing resources.
5. The system of claim 4, wherein, The interface abstraction layer includes the Deep Copy interface, the Canonicalize interface, the ADD interface, the SUB interface, the CMP interface, and the Equal interface.
6. The system of claim 3, wherein, The heterogeneous computing resources CRD are provided with CRUD interfaces through the k8s apiserver.
7. The system of claim 3, wherein, The updated scheduling policy CRD includes a built-in policy and a third-party extended policy. The built-in policy is loaded into the cache during the operation of the computing power scheduling module, and the third-party extended policy is read from the cloud-native storage ETCD module by dynamic loading during the operation of the computing power scheduling module.
Citation Information
Patent Citations
Scheduling method and system for distributed heterogeneous computing power of urban brain AI computing center
CN112532751A
Task processing method, device and equipment and computer storage medium
CN113391890A
