Availability Modeling and Analysis of a Virtualized System Using Stochastic Reward Nets

Abstract

Availability is one of the key requirements for modern networked system. Availability of a virtualized system can be modelled and analyzed using stochastic models. In our previous work, availability of a virtualized system was modeled using a hierarchical model to incorporate the detailed behavior of virtual machines (VMs)' failure and recovery with respect to the system behavior. In particular, a truncated continuous time Markov chain model was used for VM mode. In this paper, we propose to construct a stochastic reward nets (SRN) to model and analyze the availability of a virtualized system. Further, we study the effect on the availability when restrictions on the guard functions are relaxed. We use a virtualized data center (VDC) using three hosts with a multiple number of VMs on each hosting servers, and we incorporate (i) failure and recovery of hosting servers and VMs, (ii) VM high availability (HA) and (iii) VM live migration (LM). The VDCs with/without using VM HA and LM are compared in terms of capacity oriented availability (i.e., number of available VMs).