Improve Performance and Throughput of VMs for Scientific Workloads in a Cloud Environment

Abstract

Today, the latest computing paradigm that can fulfill the rapid growing demand of computational power for scientific workloads is known as cloud computing. Nowadays, scientific community is also interested to take advantage of cloud technology. To run the scientific workloads in the cloud environment required provisioning of logical resources known as virtual machines (VMs). However, there are some significant problems related to performance and throughput of virtual machines. This paper uses the cloud technology for scientific workloads and addresses the issues related to performance and throughput and efficiently provisioned of VMs for scientific workloads in a cloud computing environment. We evaluated the performance and throughput by using HEPSCPEC06 benchmark suite. Our proposed solution combines the four basic techniques to minimize the impact of virtualization and improve the overall performance and throughput of a virtual machine. The results shows that the maximum performance and throughput of virtual machines can be achieved by enabling the hyper-threading, properly assign the number of CPU cores, isolation of cores and pinning of vCPUs cores.