Virtual Cluster Placement in Datacenters

Large datacenters hosting multiple virtualized servers, storage, and networked clusters are gainingpopularity due to the economies of using shared commodity hardware. To maximize these benefits, it is necessaryto obtain high consolidation ratios by optimizing the usageof compute, storage and network resources. The optimalplacement of VMs is difficult since it must deal with multiple client resource requirements (e.g. vCPUs, memory, storage, IOPS, network bandwidth), and a heterogeneousphysical infrastructure with islands of storage and networkconnectivity. We formulate ILP models for VM placement involvingmultiple resources in a number of scenarios. We presenta novel technique for the co-location of communicatingclusters of VMs that significantly reduces the number ofmodel variables. The models developed can be used foron-line placement of incoming VM clusters in a runningdata center.