On Enabling Energy-efficiency in Geographically Distributed Data Centers through Data Deduplication

Abstract

Cloud Service Providers (CSPs) are increasingly adopting geographically dispersed data centers (DCs) to cater to growing user demands while minimizing latency. However, this distributed architecture poses significant energy consumption challenges. While research on cost-reduction techniques for geographically distributed DCs exists, a clear gap remains in exploring cost savings during workload processing, particularly when incorporating data deduplication techniques. This paper proposes Data Deduplication Geographical Load Balancing (DDGLB), a novel approach designed to reduce operating costs in geographically distributed data centers. DDGLB leverages a data deduplication algorithm, such as Knuth-Morris-Pratt (KMP), to compare incoming workloads with those previously processed within a specific DC. If a match is found, the global load balancer (GLB) refrains from forwarding the duplicate workload for processing. Instead, it returns the pre-computed results, achieving significant energy savings. Conversely, if no match is found, the GLB allocates the new user request to a specific DC for processing. This allocation considers various optimization parameters, including the availability of historical workload traces, on-site renewable energy sources, dynamic energy prices, and the number of active servers at each location. Through comprehensive numerical evaluations, this study demonstrates that DDGLB can significantly reduce operating costs for geographically distributed DCs compared to existing user request distribution strategies.