Abstract:

In realistic scenarios of cloud computing, performance of a cloud service is indeed a random variable due to random resource failures. To make more precise evaluation, the random change of performance caused by resource failures and subsequent recovery should be captured. In this paper, we consider a cloud system that has a hierarchical failure recovery mechanism consisting of three typical kinds of repair. A theoretical modeling approach is presented to evaluate the performance metric that captures both service time of virtual machines (VM) and parse time of the cloud controller (CC) to ensure high fidelity. Numerical examples are illustrated.