An Container Elastic Autoscaling Strategy Based Adaptive Integrated Resource Forecast

The built-in elastic autoscaling strategy of Kubernetes, a container cloud orchestration and management system, balances the relationship between application service quality and cluster resource usage by regularly monitoring a single resource and comparing it with a set threshold to determine whether to scale up or down. However, the built-in strategy has the problems of incomplete evaluation of a single resource index, response delay and scaling jitter. To address the above problems, this paper uses an adaptive weighting algorithm to obtain a comprehensive prediction index based on long-range time multi-resource prediction sequences. This index is not only more comprehensive than a single index, but also measures the resource usage in the future time period, so that the adaptive comprehensive prediction index has the ability to more comprehensively measure the changes in application workload in the future time period. The experimental results show that the autoscaling strategy based on adaptive integrated resource forecast can effectively solve the problems of regulation lag and scaling jitter.