Towards Optimal Electric Demand Management for Internet Data Centers

Electricity cost is becoming a major portion of Internet data center (IDC)´s operation cost and large-scale IDCs are becoming important consumers of regional electricity markets. IDC´s energy efficiency is gaining more attention by data center operators and electricity market operators. Effective IDC electric demand management solutions are eagerly sought by all stakeholders. In this paper, a mixed-integer programming model based IDC electric demand management solution is proposed, which integrates both the impacts of locational marginal electricity prices and power management capability of IDC itself. Dynamic voltage/frequency scaling of individual server, cluster server ON/OFF scheduling, and dynamic workload dispatching are optimized while complying with all the IDC system-wide and individual heterogeneous servers´ operation constraints according to the IDC applications´ temporal variant workload. Reduced electricity cost can be achieved together with guaranteed QoS requirement and reliability consideration by using the proposed model. World Cup ´98 data is utilized to evaluate the effectiveness of the proposed solution. According to the experimental evaluation, electricity cost could be cut by more than 20% in a peak workload period and by more than 80% in a light workload period. Besides, more than 6% electricity cost could be cut by considering the impact of electricity price difference. Experimental results also reveal that higher QoS requirement and reliability consideration could result in higher electricity cost.