Performance analysis of a multi-tier video on demand service on virtualization platforms

Abstract

Cloud computing technology, especially virtualization is employed in many data centers nowadays. The key concept of virtualization concerns elastic or scalable infrastructure. This concept can be implemented by exploiting multi-tier web applications and hypervisors which are virtual machine management software. Xen hypervisor has presented its Version 4.4 in 2014. In this paper, we present the performance analysis comparison between Xen para-virtualization and KVM full virtualization on the case study of open source video streaming called Cumulusclips, a YouTube-like system. This investigation involves real workload mp4 video streaming on 200 clients' browser, running 3 experiments, including large video files (∼3 GB), small video files (∼120 MB) and random-size video files (the ratio of large and small video files is 25%/75%). The requests size results show that Xen para-virtualization can serve all requests better than KVM full virtualization and use less resource. Xen's performance is dropped when CPU usage is 100% in Experiment 1 (large files). In Experiments 2 (small files) and 3 (random-size files), Xen's CPU usage is under 10%, but KVM's CPU usage is over 50%. The requests size results of Xen and KVM are equal in Experiment 2, but Xen has the maximum throughput about a half (51%) of KVM's. Therefore, we can conclude that Xen para-virtualization has better performance than KVM full virtualization on multi-tier video streaming system.