Research On Container Scaling Mechanism Based On Three-way Decision

Cloud computing is a commercialized model that provides computing resources as a service for payment.Among them,how to reasonably and efficiently allocate resources based on the cluster load and user service level is an essential content of cloud computing research.There are also problems in the field such as untimely expansion and reduction of capacity and imbalance in resource use.Such issues will significantly affect the service quality of the cluster.The problem of container expansion and contraction can be divided into two types: horizontal expansion and contraction and vertical expansion and contraction.Horizontal expansion and contraction refers to increasing the number of resources providing services,while vertical expansion and contraction refers to improving the performance of a single service resource.These methods are all effective ways to improve the efficiency of container expansion and reduction.Therefore,if you analyze the cluster's load,CPU utilization,and other attributes in the container expansion and shrinkage field,and propose new horizontal and vertical container expansion and reduction algorithms,it will be beneficial to improve the cluster's service quality and user satisfaction.This article addresses a series of "three" phenomena in cloud computing,such as the common number of copies in the container is three,the scheduling mode of the container can be divided into "centralized scheduling","two-tier scheduling" and "shared state scheduling",and the container scaling capacity also needs to be distinguished: "scaling capacity","scaling capacity" and "jitter",so this paper introduces three decisions into the container scaling capacity domain,combined with the idea of three decisions,two container scaling capacity optimization methods are given.(1)Horizontal Pod Autoscaler(HPA),which is the mainstream standard Kubernetes container scaling algorithm in the container scheduling industry,cannot cope with drastic request load changes,and the expansion and shrinkage are not timely,resulting in poor user experience and even cluster collapse.In this paper,propose a container level scaling capacity algorithm based on three-way decision(Tri Horizontal Pod Autoscaler,THPA).The THPA algorithm first records the CPU utilization of the container cluster into a queue,if there are currently high load requests,CPU utilization exceeds the set threshold value,it is determined that there are high load requests to arrive,start expansion operation,for the number of container expansion,this paper sets two thresholds A,B,through these two thresholds and then further delineate the container expansion range.Similarly,simulation experiments show that the algorithm can effectively improve the container's scaling capacity efficiency,while effectively removing scaling capacity jitter,ensuring the SLA of the service during the peak load period,improving the cluster service quality and ensuring cluster security.(2)Kubernetes vertical scaling is implemented by the container vertical scaling algorithm(Vertical Pod Autoscaler,VPA),there are a variety of user requests in the current cloud computing,but VPA still use a fixed scaling mechanism to deal with container scaling problems,which can no longer meet the development needs of container vertical scaling.Therefore,this paper proposes a triple decision-based container vertical scaling capacity algorithm(Tri Vertical Pod Autoscaler,TVPA)to solve the above problem.The TVPA algorithm statistics the historical CPU utilization of a container,while borrowing the idea of the three decisions,the CPU cores of the container and the memory limit of the expansion process is processed in three parts,if the CPU utilization of a container in the statistical cycle,there are enough data greater than the set expansion threshold,then the container's CPU and memory limit will be moderately expanded by a specific algorithm.Similarly,on this basis,the strength of the expansion capacity will be further divided into three parts,so as to meet the container expansion capacity needs in different contexts...