Versatile workload-aware power management performability analysis of server virtualized systems

Abstract

Abstract The widespread integration of virtualization technologies in data centers has enabled in the last few years several benefits in terms of operating costs and flexibility. These benefits maybe boosted through join optimization of power management (PM) and dependability for virtualized systems. This indeed involves developing appropriate models to better understand their performability behavior whenever they are exposed to predictable (e.g. rejuvenation) and unpredictable breakdowns. We propose in this paper a performability analysis of server virtualized systems (SVSs) using a workload-aware PM mechanism based on non-Markovian Stochastic Reward Nets (SRNs) modeling approach. This analysis investigates interactions and correlations between several modules involving workload-aware PM mechanism, dynamic speed scaling processing, virtual machine (VM) and virtual machine monitor (VMM) both subject to software aging, failure and rejuvenation. We show through numerical results, using quantitative and qualitative metrics, how performance, power usage and efficiency are impacted by workload-aware PM mechanism. We show also how judicious choice of tunable attribute (i.e. Timeout ) of the proposed PM mechanism with respect to workload can lead to a good power-performance trade-off.