Unified power and performance analysis of cloud computing infrastructure using stochastic reward nets

In this paper, power and performance of cloud infrastructures are analytically modeled and numerically evaluated using stochastic reward nets (SRNs). The main objective of the proposed model is to analyze the impact of the temperature and network traffic on power and performance of cloud computing systems, which has been ignored in most of previously presented approaches. In the proposed model, incoming tasks are divided into two categories, cold and hot, demonstrating different power consumption patterns, that can fail due to bandwidth constraints and congestion. By applying the proposed model, different resource allocation scenarios can be evaluated and compared, in terms of power and performance, when hot and cold tasks are assigned to virtual machines and the network congestion is taken into consideration. The results obtained from the proposed model, cross-validated with the CloudSim framework, show that the scenario which uses the minimum number of racks and the maximum number of servers is optimal in terms of both the power and performance measures.