Maximum elastic scheduling of virtual machines in general graph cloud data center networks

Abstract

The assignment of task resources has drawn a lot of attention in the applications of cloud-based data center networks (DCNs). In this work, we provide a novel method that evaluates the maximum admissible load (MAL) given a set of virtual machines (VMs) in physical machines (PMs). This is the first study that considers the general case of DCNs with no constraints on their topologies, contrary to previous research. We prove that this problem can be reduced to a single-source, multiple-sink maximum flow problem. In addition, we find the optimal load assignment that supports the maximum uniform growth rate without resorting to task reassignment to PMs. This is called the maximum elastic scheduling problem. A new effective strategy is devised to solve the two problems. We demonstrate the efficiency of its performance using an extensive simulation.