A Flexible Model for Predicting Workloads of Massive Microservices

Deploying microservices into containers allows for dynamic adjustment of the resources required by each microservice based on predicted future workload. Typically, a cluster hosts a large number of different microservices. Due to frequent scheduling, most microservice containers have a short lifespan. Microservice workload prediction can help resource management. Microservice poses a challenge to workload prediction: how to flexibly predict massive short microservice workloads. In this paper, we propose a prediction model, DKM, which generates the Dynamic Kernel for each microservice, so as to flexibly predict massive diverse microservice workloads. DKM can learn from massive microservice workloads to build its own Kernel Dictionary, allowing it to generate the Dynamic Kernel for different microservices by selecting appropriate kernels from the dictionary. Experiments demonstrate that DKM performs well in workload prediction.