A multi-resource indicator Kubernetes scheduling method based on delay factor

Abstract

The present invention discloses a multi-resource indicator Kubernetes scheduling method based on delay factor, which is characterized in that the priority of cloud computing tasks is evaluated according to specific evaluation criteria based on the resource request of cloud computing tasks; scheduling delay is set for cloud computing tasks according to the priority of cloud computing tasks and the status of computing resources; cloud computing tasks are inserted into a delay sequence according to the scheduling delay of cloud computing tasks; multi-resource indicator Node matching with dynamic weights is performed according to the resource occupancy rate of cloud computing tasks and the idle status of Node resources, and cloud computing tasks are allocated to the optimal Node according to the matching degree. The present invention realizes efficient resource scheduling for cloud computing tasks of different types of resources through multi-resource indicator Node matching with delay factor and dynamic weight, ensures timely response of cloud computing tasks, improves the utilization rate of cloud computing resources, and realizes load balancing between various resources.

Description

Technical Field

[0001] The present invention belongs to the technical field of cloud computing, and in particular relates to a multi-resource indicator Kubernetes scheduling method based on a delay factor. Background Art

[0002] Cloud computing technology is the trend of today's wave and has a huge market prospect. Cloud computing is penetrating into every corner of life and triggering a complete transformation of various industries. Kubernetes is the most widely used container cloud orchestration management system in the field of cloud computing and an important supporting technology for cloud computing services.

[0003] The traditional Kubernetes scheduling method schedules nodes based on two resource indicators: node memory and CPU. However, the resources involved in cloud computing tasks include not only CPU and memory, but also bandwidth, disk and GPU. The traditional scheduling method cannot meet the resource scheduling requirements of multiple resource indicators in cloud computing scenarios. In addition, the resource request response of cloud computing tasks is time-series, while the traditional Kubernetes scheduling belongs to static resource scheduling, which does not fully consider the resource requests in the future, which may lead to many problems such as the failure of computing tasks to be responded to in time, low resource utilization and unbalanced resource load. Summary of the invention

[0004] In order to overcome the shortcomings of traditional Kubernetes scheduling methods, the present invention provides a multi-resource indicator Kubernetes scheduling method based on delay factors. The method fully considers resource requests in the future, sets scheduling delays for computing tasks according to task priorities and computing resource states, and performs multi-resource indicator Node matching with dynamic weights for cloud computing tasks with multiple resource indicators, thereby realizing efficient resource scheduling of cloud computing tasks, ensuring timely response of cloud computing tasks, improving the utilization rate of cloud computing resources, and realizing load balancing among resources.

[0005] To achieve the above object, the technical solution of the present invention is:

[0006] A. Priority evaluation of cloud computing tasks: Calculate the required resource ratio s based on the CPU, memory, bandwidth, disk and GPU resource requests of cloud computing tasks. i , according to s i Prioritize cloud computing tasks according to specific evaluation criteria i Make an evaluation.

[0007] B. Cloud computing task scheduling delay setting: Set the scheduling delay for cloud computing tasks according to the cloud computing task priority and computing resource status. Set a longer scheduling delay for low-priority tasks, and a shorter delay or no delay for high-priority tasks. Set a longer scheduling delay when computing resources are tight, and set a shorter delay or no delay when computer resources are abundant.

[0008] C. Cloud computing task delay: insert the cloud computing task into the delay sequence according to the scheduling delay of the cloud computing task, and wait for the delay to end before scheduling the cloud computing task Node.

[0009] D. Node scheduling for cloud computing tasks: Node pre-selection is performed based on the resource requests of cloud computing tasks, and nodes that do not meet the requirements are excluded; dynamic weighted multi-resource indicator node matching is performed among the candidate nodes, where the resource occupancy rate of the cloud computing task determines the dynamic weight of the matching function, and the idle status of each node resource determines the degree of matching; cloud computing tasks are assigned to the optimal node based on the degree of matching. If the assignment fails, the node re-enters the delay queue.

[0010] The beneficial effects of the present invention are: fully considering resource requests in the future, realizing delayed scheduling of computing tasks based on task priority and computing resource status, realizing multi-resource indicator Node matching with dynamic weights for cloud computing tasks with multiple resource indicators, realizing efficient resource scheduling of cloud computing tasks, ensuring timely response of cloud computing tasks, improving cloud computing resource utilization, and realizing load balancing among resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 This is the model structure diagram of the present invention DETAILED DESCRIPTION

[0012] A. Priority evaluation of cloud computing tasks: Evaluate the priority of cloud computing tasks based on their specific resource requests. Calculate the required resource ratio s based on the resource requests of cloud computing tasks. i ,s i The calculation formula is as follows:

[0013] s i =(s cpu +s mem +s bandwidth +s disk +s gpu ) / 5,

[0014]

[0015]

[0016]

[0017]

[0018]

[0019] Among them, cpu i 、mem i 、bandwidth i 、disk i 、gpu i They represent the computing resources requested by cloud computing task i. average 、mem average 、bandwidth average 、disk average 、gpu average Respectively represent the average value of computing resources of all Nodes. cpu 、s mem 、s bandwidth 、s disk 、s gpu They respectively represent the resource proportions of computing resources required for cloud computing task i.

[0020] According to s i Prioritize cloud computing tasks according to specific evaluation criteria i Make an evaluation, evaluation criteria: i When <5%, p i 1; 5%≤s i When <20%, ​​p i 2; 20%≤s i When <35%, p i 3; 35%≤s i When <50%, p i is 4;s i When ≥50%, p i is 5; p i The larger the value, the higher the task priority.

[0021] B. Cloud computing task scheduling delay setting: Due to the timing of cloud computing task requests, the following two cases are considered: R1 is a cloud computing resource task request with a priority of 3, and R2 is a cloud computing resource task request with a priority of 4.

[0022] (1) When computing resources are tight, R1 and R2 are matched. The matching node that meets R2's requirements is Node1, and the matching nodes that meet R1's requirements are Node1 and Node2. The corresponding matching degrees are m1 and m2 (m1>m2). If R1 arrives at time T, the system assigns it to Node1. Later, R2 arrives at time T+t. Node1 has insufficient resources and R2 cannot get a response.

[0023] (2) Under the same circumstances, R1 and R2 are matched. The best matching nodes of R1 and R2 are both Node1, and the corresponding matching degrees are m3 and m4 (m3<m4). The second best matching nodes of R1 and R2 are both Node2, and the corresponding matching degrees are m5 and m6 (m5>m6). If R1 arrives at time T, the system will assign it to Node1. If R2 arrives at time T+t, the system will assign it to Node2.

[0024] It is obvious that traditional allocation schemes have various drawbacks under the timing of cloud computing. Not only can they not guarantee the timely response of cloud computing tasks, but they also cannot meet the efficient resource utilization of the service cluster and the load balancing between resources. To avoid such problems, a scheduling delay is set for cloud computing tasks according to the priority of cloud computing tasks and the status of computing resources. For low-priority tasks, a longer scheduling delay is set, and for high-priority tasks, a shorter delay or no delay is set. When computing resources are tight, a longer scheduling delay is set, and when computer resources are abundant, a shorter delay or no delay is set. The delay calculation formula is as follows:

[0025]

[0026]

[0027] where t i The delay set for task i, t is a unit time, q is the total resource utilization, is the total occupancy ratio of each computing resource.

[0028] C. Cloud computing task delay: insert the cloud computing task into the delay sequence according to the scheduling delay of the cloud computing task, and wait for the delay to end before scheduling the cloud computing task Node.

[0029] D. Cloud computing task node scheduling: pre-select nodes according to cloud computing task resource requests, and exclude nodes that do not meet the requirements; perform dynamic weighted multi-resource indicator node matching among candidate nodes, where the resource occupancy rate of the cloud computing task determines the dynamic weight of the matching function, and the idle state of each node resource determines the matching degree. The matching function calculation formula is as follows:

[0030]

[0031] Among them, w cpu 、w mem 、w bandwidth 、w disk 、w gpu is the dynamic weight of the five resources, w cpu :w mem :wbandwidth :w disk :w gpu =s cpu :smem:s bandwidth :s disk :s gpu and It is the idle ratio of each computing resource in Node.

[0032] The cloud computing task is assigned to the optimal Node according to the matching degree. If the assignment fails, it will re-enter the delay queue.

[0033] The above is only a preferred embodiment of the present invention. Any technician familiar with the profession may use the above technical solution to modify or change it into an equivalent example with equivalent changes. Any simple modification, change or modification of the above embodiment based on the technical solution of the invention without departing from the content of the technical solution of the present invention shall fall within the protection scope of the technical solution of the invention.

Claims

1. A multi-resource indicator Kubernetes scheduling method based on delay factor, characterized in that: The following steps are involved: A. Priority evaluation of cloud computing tasks: Calculate the required resource ratio s based on the CPU, memory, bandwidth, disk and GPU resource requests of the cloud computing task i ,s i The calculation formula is as follows: s i =(s cpu +s mem +s bandwidth +s disk +s gpu ) / 5, Among them, cpu i 、mem i 、bandwidth i 、disk i 、gpu i They represent the computing resources requested by cloud computing task i, cpu average 、mem average 、bandwidth average 、disk average 、gpu average Respectively represent the average value of computing resources of all Nodes, s cpu 、s mem 、s bandwidth 、s disk 、s gpu They represent the resource proportions of the computing resources required for cloud computing task i respectively; According to s i Prioritize cloud computing tasks according to specific evaluation criteria i Make an evaluation, evaluation criteria: i <5%, p i 1; 5%≤s i <20%, p i 2; 20%≤s i <35%, p i 3; 35%≤s i <50%, p i is 4;s i When ≥50%, p i is 5; p i The larger it is, the higher the task priority; B. Cloud computing task scheduling delay setting: Setting a scheduling delay for cloud computing tasks according to the cloud computing task priority and computing resource status, setting a longer scheduling delay for low-priority tasks, setting a shorter delay or no delay for high-priority tasks, setting a longer scheduling delay when computing resources are tight, and setting a shorter delay or no delay when computer resources are abundant; C. Cloud computing task delay: According to the scheduling delay of the cloud computing task, the cloud computing task is inserted into the delay sequence, and the cloud computing task Node is scheduled after the delay ends. The delay calculation formula is as follows: where t i The delay set for task i, t is a unit time, q is the total resource utilization, is the total occupancy ratio of each computing resource; D. Cloud computing task Node scheduling: Node pre-selection is performed based on the resource request of the cloud computing task, and nodes that do not meet the requirements are excluded; dynamic weighted multi-resource indicator node matching is performed among the candidate nodes, where the resource occupancy rate of the cloud computing task determines the dynamic weight of the matching function, and the idle state of each node resource determines the degree of matching; the cloud computing task is assigned to the optimal node based on the matching degree. If the assignment fails, it will re-enter the delay queue.

Images