A Performance Study of Hardware Impact on Full Virtualization for Server Consolidation in Cloud Environment

Underutilization of hardware resources has always been a problem in single workload driven traditional OS environment. To improve resource utilization, virtualization of multiple VMs and workloads onto the same host with the aid of Hypervisor has been the recent trend. Its use cases such as server consolidation, live migration, performance isolation and on-demand server provisioning make it as a heart part of enterprise application cloud. Cloud is an on-demand, service provisioning technology, where performance plays a vital role for user acceptance. There are numerous virtualization technologies are available from full virtualization to paravirtualization, each has its strength and weakness. As performance study is an ongoing pursuit, hardware and software development getting matured day by day, it is desirable to do this sort of performance study in regular interval that often sheds new light on aspects of a work not fully explored in the previous publication. Hence, this paper focus performance behaviours of various full virtualization models such as hosted (VirtualBox) and bare metal (KVM) virtualization using variety of benchmarks from micro, macro and application level in the cloud environment. We compare both virtualization solutions with a base system in terms of application performance, resource consumption, low-level system metrics like context switch, process creation, interprocess communication latency and virtualization-specific metrics like virtualization layer consumption. Experimental results yield that VirtualBox outperforms KVM in CPU and thread level parallelism and KVM outperforms in all other cases. Both are very reluctantly accepted for disk usages comparing with native system.