Queueing analysis of migration of virtual machines

Abstract

Live migration is a prerequisite feature of virtualization that allows transfer of a working virtual machine from one data center to another. It is a useful tool for optimization of resources and is one of the issues of research. The work [22] proposes an analytical model based approach for quality evaluation of cloud by considering the rejection probability, expected request completion time, system overhead as key metrics. In this paper, we are exploring only rejection probability of jobs. The work in [22] has been done in two phases and M/M/1/K queueing model has been applied to the first phase of the work in order to find the rejection probability of jobs. We are considering the variable buffer sizes and different distribution models. To study the impact of changing buffer sizes on the rejection probability of the jobs is the main point of concern here. We have proposed application of M/G/1/∞, M/G/1/K models to it. And, in order to validate the correctness of the proposed model, we simulate the data and the graphs are drawn in Matlab showing the comparison of the proposed model with work presented in [22]. It is observed that in case of M/M/1/K, with an increase in the request arrival rate, the rejection probability of jobs increases, but if we change the model to M/G/1/∞, the rejection probability of jobs decreases. With an increase in execution rate, the rejection rate of jobs decreases, but if we change [22] to M/G/1/∞, then the rejection probability decreases more as compared to it. Hence, changing buffer size proves to be gainful. And the changing of queueing model is advantageous as it leads to decrease in the rejection probability of the jobs. So, we can say that general distribution queueing model is more effective in migration as compared to exponential distribution model as it leads to a decrease in rejection rate of jobs.