On a fault-tolerant resource allocation scheme for revenue maximization in data centers

摘要

2011 has been the “Year of the Cloud” as organizations around the world have started migrating their services and applications to large-scale data center infrastructures. A large-scale data center experiences random failures of several hardware components every day. These hardware failures present challenging issues in providing reliable service to the end users. The notion of fault domains captures the effect of single hardware failures in data centers. In this paper, we use the concept of fault domains and investigate the problem of revenue maximization in fault-tolerant resource allocation in large data centers. We provide a novel formulation of the problem and prove that this problem is NP-complete. We provide optimal solution technique through Integer Linear Program formulation and present an efficient heuristic that produces near-optimal solution in a fraction of time required to compute the optimal. Through extensive experimentation, we prove the efficacy of our heuristics.