Unavailability-aware allocation of backup resources considering failures of virtual and physical machines

Abstract

Cloud service providers widely use virtual machines and bring various benefits for customers and themselves. The unavailability of virtual machines is a crucial performance indicator for cloud service quality. However, no previous works studied the unavailability-aware backup allocation considering the failures of both virtual and physical machines. This paper proposes a backup allocation model considering these failures to minimize the maximum unavailability among virtual machines. We use a Markov chain and analyze the state transitions of the proposed model. We establish how to analytically compute the unavailability in the proposed model by solving the equilibrium-state equations formulated from the state transitions. We introduce a heuristic algorithm to solve the optimization problem of backup allocation. Numerical results show that the proposed model reduces the maximum unavailability by 7.5–28.2% compared to baseline models. Our study establishes a fundamental analytical way to calculate unavailability in the proposed model, which handles the failures of both virtual and physical machines.