Containers Scheduling Consolidation Approach for Cloud Computing

Abstract

Containers are increasingly gaining popularity and are going to be a major deployment model in cloud computing. However, consolidation technique is also used extensively in the cloud context to optimize resources utilization and reduce the power consumption. In this paper, we present a new containers scheduling consolidation approach for cloud computing environment based on a machine learning technique. Our approach is proposed to address the problem of a company that aims to adapt dynamically the number of active nodes to reduce the power consumption when several containers are submitted online each day by their users. In our context, the frequency of containers submission varies within one hour. However, for each hour, the submission frequency is essentially the same each day. The principle of our approach consists into applying a machine learning technique to detect, from a previous containers submission historical, three submission periods (high, medium and low). Each submission period represents a time slot of one day. For instance, the high submission period represents the slot time where the number of submitted containers is the highest compared to other periods. Then, according to the submission periods slot time, our approach dynamically adapts the number of active nodes that must be used to execute each new submitted container. Our proposed consolidation approach is implemented inside Docker Swarmkit which is a well-known container scheduler framework developed by Docker. Experiments demonstrate the potential of our approach under different scenarios.