Research On Container Scaling Optimization Based On Gray-Markov

With the swift development and extensive utilization of cloud computing technology,container technology has emerged as an indispensable part of cloud environment’s resource scheduling and cluster management.In contrast to conventional virtual machine-based technologies,container technology demonstrates higher resource utilization and faster resource allocation speed.Containers can be swiftly scaled up to increase available resources and ensure successful task completion when resources are inadequate.Similarly,when resources are surplus,they should be reclaimed promptly to release more assignable resources to other containers.These functionalities make container technology extensively used.However,resource scheduling in a rational and efficient manner to meet the correlation between cluster load and user service requests remains a crucial issue in current cloud computing applications.In cloud environments,an abundance of task requests may appear at any time,but containers cannot always respond on time to meet task requirements,causing issues such as long task request response times and an increased rate of Service Level Agreement(SLA)violation,significantly impacting service quality and user experience.If containers directly reclaim resources when the task request volume suddenly decreases,they will slow down again when facing sudden bursts of requests.Frequent scaling up and down not only fails to improve resource utilization but also keeps some resources in the scheduling and allocation state,increasing server power consumption.In container clouds,traditional automatic scaling mechanisms are categorized into horizontal and vertical automatic scaling mechanisms.The horizontal automatic scaling mechanism increases or decreases the number of containers to achieve scaling up or down,while the vertical automatic scaling mechanism increases or decreases the total amount of resources of a single container to achieve scaling up or down.The automatic scaling mechanism scales up or down based on the load situation to ensure that cloud services can operate stably and normally.However,in environments with significant load fluctuations and unpredictability,traditional scaling mechanisms cannot guarantee the normal operation of cloud services.This paper proposes an automatic scaling optimization strategy based on grey Markov models to address the problem of Kubernetes automatic scaling policies being unable to timely scale up or down in environments with large load fluctuations and strong unpredictability.The proposed strategy combines reactive scaling policies and predictive scaling policies.Specifically,this paper introduces two strategies,GMA-HPA(Grey Markov-Horizontal Pod Autoscaler)and GMA-VPA(Grey Markov-Vertical Pod Autoscaler).(1)The GMA-HPA strategy utilizes a Grey Markov model to predict the short-term load fluctuations of containers by analyzing historical data.The next time-step’s load prediction is made based on the historical load sequence,and the prediction result is adjusted using the Markov model to reduce prediction errors.Finally,containers are proactively scaled up based on the predicted result to meet task requirements,thereby reducing SLA violations and improving the service quality and response speed of containers.This optimization algorithm effectively achieves rapid scaling to meet sudden task requests and slow scaling to release resources,reducing cloud task violations and lowering the SLA violation rate.(2)This passage describes the GMA-VPA strategy,which addresses the issue of the fixed scaling mechanism of the vertical automatic scaling algorithm in Kubernetes,where adding or reclaiming resources to a single container can no longer guarantee the stable operation of the container cluster.To solve this problem,a container vertical scaling strategy based on the grey Markov chain model is proposed.The strategy utilizes the grey Markov model to predict the cluster load for the next time step based on historical time series data and adjusts the resource utilization of individual containers accordingly.The optimized algorithm achieves both rapid scaling up and slow scaling down of individual containers and effectively reduces SLA violations to ensure the normal and stable operation of cloud tasks.