A model for evaluating the economics of cloud federation

摘要

We consider the problem of formation of an economically sustainable computational resource federation by Cloud Service Providers (CSPs). The federation aims at providing improved quality of service, expressed in terms of average total delay per job provided to clients served by the CSPs. Each CSP is modeled as an M/M/1 queue. The queue may serve requests coming from that CSP's own client, as well as requests coming from clients of other CSPs. Our model includes all salient factors to evaluate the economics of such a federation: we model the energy consumption cost of each CSP as a function of its resource utilization factor, and we model the CSP's revenue by a delay-dependent pricing function according to which each CSP charges its clients. We propose a model for the formation and evaluation of a cloud federation according to which each CSP may transfer a portion of the requested workload from its clients to other CSPs within the federation. A federation policy is specified by the portions of workload transferred from each CSP to other CSPs. We formulate the problem of finding the federation policy that maximizes the total profit (revenue minus cost) of CSPs and we also deal with the incentives of individual CSPs. Finally, we conduct several experiments under different setups. The numerical results show that according to our model CSPs can maximize the total profit of the federation and also achieve a nearly optimal QoS.