Balanced Graph Partitioning based I/O Intensive Virtual Cluster Placement in Cloud Data Center

Abstract

Distributed application deployment in a cloud environment has emerged as a new area of research in cloud computing. I/O intensive virtual machines form a virtual cluster, which requires a large volume of data transfer for communication among themselves. Cloud service providers manage huge infrastructural resources using standard hierarchical network topologies, like FatTree, VL2, etc. To achieve high throughput communication cost needs to be reduced. In this regard, efficient allocation of the cooperating virtual machines may reduce communication costs and delay without compromising total energy consumption by the active hosts of the data center. However, the objectives are conflicting with each other. This research proposes a novel balanced graph partitioning-based heuristic that reduces communication costs by allocating the cooperating VMs at proximity and energy consumption by consolidating the virtual machines in a minimum number of active hosts for reducing wastage of the active hosts, which in turn reduces the energy consumption of the data center.