LAUSR

Many critical e-commerce and financial services running on distributed data centers require high availability. Recent surveys show that the frequency and duration of failures (either partial or complete) at a data center are increasing over the years. However, most of the literature does not address fault-tolerance in data centers. In this paper, we address the problem of load balancing in fault-tolerant data centers while minimizing the operating cost (due to energy consumption). We formally model the problem using linear optimization and propose a distributed two-stage algorithm to solve it. In the first stage, the proposed workload shifting algorithm distributes the load on a failed data center across the remaining ones with a marginal increase in the operating cost. Subsequently, the min-cost network flow model is used to derive a request mapping policy, where the quality of service (QoS) requirements are considered using a set of constraints on the delay. We evaluate the proposed algorithms using real-world data for the demand and the energy consumption cost (both brown and green energy). Results show that the proposed algorithm has low computational complexity, yet exactly gives the cost obtained by the global optimal solution.