Cloud computing infrastructures, consisting of myriad central processing units (CPUs), storage devices, and network connections, depend heavily on virtualized components. Virtualized compute, network, and input/output (I/O) resources can be located almost anywhere within a cloud infrastructure, necessitating optimal virtual machine (VM) placements in order to reduce power consumption and to avoid performance bottlenecks.

This paper presents a novel algorithm for the placement of groups of VMs. It focuses primarily on network flow (bandwidth) optimization, and also takes into account physical server and network switch energy consumption. The algorithm aims to minimize equipment usage, using a fat tree topology that accounts for higher bandwidth at the network endpoints (VMs).

The authors evaluated their algorithm’s utility and scalability via simulation on some rather limited and older Intel servers, although in comparison to other simulation-based VM placement algorithms it appears to perform well. Testing of the algorithm in actual cloud implementations would be of value. Additionally, the authors’ assumptions about fixed and failure-free network topologies, uniform server types, and VM sizes and arrival rates demand further testing, especially in light of the rapid evolution of cloud computing and I/O technologies. Modern cloud infrastructures now include a variety of compute node architectures, solid-state and memory-based storage, and high-speed server interconnects such as InfiniBand, all of which make optimal VM placement a continually moving target.

Such VM placement research is of critical interest to cloud computing providers who seek efficient, cost-effective, and scalable solutions.