Towards Maximal Service Profit in Geo-Distributed Clouds

With the proliferation of globally-distributed services and the quick growth of user requests for inter-datacenter bandwidth, cloud providers have to lease a good deal of bandwidth from Internet service providers to satisfy the user demands. Neither maximizing the service revenue nor minimizing the service cost can bring the maximal service profit to cloud providers. The diversity of user requests and the large unit of inter-datacenter bandwidth further increase the difficulty of scheduling user requests. In this paper, we propose a cloud operational model to help cloud providers to make more service profit by properly selecting requests to serve rather than serving all user requests. We formulate the problem of service profit maximization and prove its NP-hardness. Considering the complicated coupling between maximizing revenue and minimizing cost, we propose a framework, Metis, for the efficient scheduling of user requests over inter-datacenter networks to maximize the service profit for cloud providers. Metis is formed with the alternate operations of two algorithms derived from randomized rounding techniques and Chernoff-Hoeffding bound. We prove that they can provide the guarantees on approximation ratios. Our extensive evaluations demonstrate that Metis can achieve more than 1.3x the service profits of existing solutions.