基于Kubernetes调度器的服务质量优化调度算法研究

Research and Implementation of Optimal Scheduling Algorithm for QoS Strategy based on Kubernetes Scheduler

针对现有调度算法在调度时未考虑QoS等级的影响,导致节点上不同优先等级的Pod数量不均,最终造成需要终止低优先级Pod时只有高优先级Pod存在的问题,深入分析评估了Kubernetes调度器的调度算法,利用分布式系统设计思想和技术(如MapReduce思想、Etcd存储等),设计了BalancedQosPriority算法。该算法使各QoS等级的Pod在集群中尽可能分散,促使节点上各个QoS等级的Pod尽可能梯度存在,以避免无低优先级Pod可终止的窘境,从而提高集群在突发负载时的可靠性。对算法进行性能评估,结果表明,BalancedQosPriority算法在解决上述问题的同时未造成明显的性能损失。

This thesis provides an in-depth analysis of the scheduling algorithm for the Kubernetes scheduler which focuses on the issue of ignoring the impact of QoS level in the application of current scheduling algorithm.This ignorance makes pods with different priorities and unbalanced numbers,which will ultimately cause low priority pods to be terminated and only retain high priority pods.The design of BalancedQosPriority algorithm is based on methodology together with technology from development of distributed system such as MapReduce and Etcd Storage.The reliability in handling sudden load for cluster has been enhanced by BalancedQosPriority algorithm by avoiding total termination of low priority pods.Pods with different QoS levels are scattered to the highest degree,which ensures each node has an entire gradient of QoS level for Pods.Furthermore,the result indicates by applying BalancedQosPriority algorithm,the problem of current algorithm will be solved without significant performance loss.