Energy-optimized dynamic deferral of workload for capacity provisioning in data centers

This paper explores the opportunity for energy cost saving in data centers that utilizes the flexibility from the Service Level Agreements (SLAs) and proposes a novel approach for capacity provisioning under bounded latency requirements of the workload. We investigate how many servers to keep active and how much workload to delay for energy saving while meeting latency constraints. We present an offline LP formulation for capacity provisioning by dynamic deferral and give two online algorithms to determine the capacity of the data center and the assignment of workload to servers dynamically. We prove the feasibility of the online algorithms and show that their worst case performances are bounded by constant factors with respect to the offline formulation. To the best of our knowledge, this is the first formulation for capacity provisioning in data centers considering workload deferral with bounded latency. We validate our algorithms on MapReduce workload by provisioning capacity on a Hadoop cluster and show that the algorithms actually perform much better in practice compared to the naive `follow the workload´ provisioning, resulting in 20-40% cost-savings.