Multi-granularity three-way decision scheduling optimization based on cloud cluster bursty load

In this paper, we analyze, predict and classify the load information in a multi-granular way based on the information of cloud platform cluster data released by Google 2019 to build three burst load prediction models. By analyzing the load on a single cloud machine, we predict its burst load in some future period. We combine the three-way decision theory to classify the machines into calm, intermediate, and active states based on the prediction results. We summarize the load characteristics of three machine states to establish three burst load models and propose different scheduling optimization strategies for the three models using the Colocation method to achieve the overall load balancing of cloud services.