On the Necessity of Right Optimizations for Live Migration of Virtual Machines

Abstract

The proliferation of cloud computing demands live migration of virtual machines between servers in and across the data center(s). While performing the backend computation for new paradigms such as the Internet of Things, live migrations over bandwidth constrained wide area network becomes a necessity. Pre-copy migration is the prevalent live virtual machine migration technique in use. Although work has been done in optimizing the migration time based on the nature of the applications - memory dirtying pattern, extra traffic generated, etc. - none of them rightly pointed out the effect of future dirty bitmap and bandwidth information in deciding the total migration time, downtime, and the extra traffic. In this paper, our objective is to develop an optimal algorithm which will form a baseline for these values and will tell us the minimum migration time that is possible. We consider two parameters - page dirty behaviour of the application and bandwidth availability to develop the optimal algorithm. We simulate our algorithms in CloudSim and show that a good prediction of the page dirtying pattern and bandwidth can benefit the cloud service providers to save the time and load due to migrations. This will help us in quantifying the closeness of the existing optimized algorithm results with respect to the least migration time that is possible.