A Versatile Traffic and Power Aware Performability Analysis of Server Virtualized Systems

We propose in this paper a workload-aware perform ability analysis of elastic server virtualized systems (SVS) based on non-Markovian SRN modeling approach. This analysis investigates correlations between several modules involving workload-aware power management (PM) mechanism, virtual machine (VM) and VM monitor (VMM) subject to aging, failure and rejuvenation. We show through numerical results how performance (i.e. waiting time), power usage as well as power-performance efficiency are impacted by power management attribute when using time varying workload with bursty nature. We show also how we can quantify PM attributes leading to small waiting time and a good utilization of the active power state.