Lyapunov-Based Energy Optimization for Deadline-Constrained Tasks across Geo-Distributed Data Centers

Abstract

Electricity cost is one of the key contributors to the data center's overall operational cost. Many researchers worked on reducing the overall electricity cost of the data center (DC) operations. In this paper, we examine the problem of minimizing the overall electricity cost of geographically distributed data centers while fulfilling the deadline requirement of the delay-tolerant workloads under time-varying system dynamics. One way of reducing the energy cost of the DCs is to exploit the dynamic electricity prices across time and location for scheduling workloads among geographically distributed data centers. Based on the Lyapunov Optimization technique, we present an online algorithm for scheduling delay-tolerant workloads called Strict Deadline-aware online Algorithm (SDA). A delay-tolerant workload means that the workload can be delayed up to a specified amount of time. The main aim of our work is to minimize the overall electricity cost of the operator while completing the workload within its acceptable deadline. The basic idea of the algorithm is to forward the workload arriving at each time slot to a specific data center considering the currently offered prices of electricity, delay of current workload, total unprocessed workload, and respecting other sets of constraints. Experiments were conducted considering multiple data centers using real-world workload traces and power rates. The results show that SDA outperforms existing workload distribution approaches to reduce the operator's overall power costs. Aside from lowering average power costs, the suggested algorithm completes all tasks within the deadline.