Bilkent University
Department of Computer Engineering


Energy Efficient Container Consolidation with CPU Usage Prediction in Data Centers


Gökalp Urul
MS Student
Computer Engineering Department
Bilkent University

One of the most important challenges that cloud providers need to deal with is minimizing power consumption of their data centers. Until now great amount of research have been done to solve this problem by proposing virtual machine consolidation algorithms and frameworks. However, with the recent advances in cloud computing, containers are increasingly gaining popularity to become a major deployment model in cloud environment and specifically in Platform as a Service. This article will focus on improving the energy efficiency of the physical servers for this new deployment model by proposing new algorithms to container consolidation. In this context most of the existing solution are based on aggressive container migration which resulting in unnecessary overhead and energy waste. In this work we will use different algorithms during the container allocation process to predict the short term future CPU utilization of physical servers. The use of predicted CPU utilization enables us to detect hot and cold spots before the SLA occurs and thereby resulting in less migration and less power consumption. We will evaluate our proposed algorithm through simulations on real workloads from the Planetlab datasets. As an initial work, we test our algorithm on virtual machine consolidation process with real workload.


DATE: 12 December, 2016, Monday @ 15:40