Adaptive Autoscaling On Kubernetes:optimization And Implementation

Nowadays,with the rapid development of containerization technology such as Docker.Developing and deploying massive scale internet application based on containerization technology has been quite common.While Kubernetes,which is a container-orchestration system,has become in fact the standard of this area.However,there are several defect about Autoscaling strategy in current version of Kubernetes.Firstly,Kubernetes’ decision on HPA(Horizontal Pods Autoscaling)operation based solely on static threshold algorithm,which make it hard to prepare for abrupt rise of traffic.Secondly,Kubernetes’ solution on Controlling replication jitters is by setting static calm-down window,which can resulted in latency of continuous upscaling or downscaling.Thirdly,Kubernetes’ strategy on downscaling nodes by detecting low resources usage can resulted in low efficiency in node downscaling.Based on above analysis,this article make changes below,Firstly,we put forward a model to predict future growth of traffic,which can react more quickly to sudden traffic growth in Kubernetes cluster.Secondly,by adaptively adjusting length of calm down window,we managed to relieve the impact on continuous horizontal autoscaling.Thirdly,by using new scheduling algorithm based on Pod Affinity,we managed to improve the efficiency on cluster downscaling.