Energy efficiency aware load distribution and electricity cost volatility control for cloud service providers

Abstract

This paper consider the case of a cloud service provider (CSP) who owns multiple geographically distributed data centers, with collocated sources of renewable energy. We investigate load distribution strategies to minimize electricity cost and increase renewable incorporation subject to compliance with service level agreement (SLA), considering the adverse effects of switching the servers. Our work provides some insights on the performance of different algorithms for geographical load balancing (GLB) in terms of electricity cost, renewable energy integration and number of server switching. Our proposed strategies incorporate a new way of capturing the server switching cost. We show that, instead of modeling switching cost through a linear function, the proposed technique of modeling switching cost through variance achieves a better tradeoff between some important parameters. Since the three major input parameters-electricity price, renewable energy and number of job requests-vary over time, the average cost of electricity per job request may also exhibit dramatic fluctuations. We propose to tackle this volatility by controlling the average cost of electricity per job request through leveraging contracts in the forward electricity market, and determine the optimal amount of electricity to be procured in the forward electricity market. We show that our proposed strategy substantially reduces the variance of the average cost of electricity per job and that this price risk mitigation is achieved with a decrease in the cumulative electricity cost.