An analysis of network I/O workloads in virtualized cloud environment

Nowadays, cloud computing is becoming a more and more prevalent platform to provide web services securely and conveniently in current data centers. For many well-known advantages, system virtualization turns to be a mainstream technology to build underlying infrastructures for cloud computing environment. Among a number of state-of-the-art Virtual Machine Monitors (VMMs), Xen is definitely to be a conspicuous hypervisor based VMM. However, there still lacks of in-depth understanding of the efficiency of I/O communication among multiple Virtual Machines (VMs) in both academic and industry areas, especially the burdensome network I/O workloads. In this paper, mathematical modeling methods are adopted to calculate the exchanged memory pages accurately in Xen I/O channel. According to our elaborately designed experimental evaluations, the proposed two equations could predict the number of exchanged memory pages around an average error rate of 6%, even in multiple VMs scenario. Meanwhile, we also analyzed the effects of varying MTU (Maximum Transmission Unit) size based on the theoretical analysis. Afterward, an effective optimization for network I/O performance is proposed in this paper as well.