Cutting the Electric Bill for Internet-Scale Systems

A. Qureshi, R. Weber, H. Balakrishnan, J. Guttag, B. Maggs, "Cutting the Electric Bill for Internet-Scale Systems," ACM SIGCOMM Conference, (August 2009).

This paper talked about how data center operating costs can be brought down by millions of dollars by offloading computation work at those locations where electricity costs are cheaper at that particular time. The authors argue that costs of electricity are highly variable at various places in the US and have little correlation between them. Further, the costs are highly dynamic which requires transferring computation continuously to different locations in a highly dynamic manner. Further, they introduce a concept of energy elasticity, that is the degree to which the energy consumed by a cluster depends on the load placed on it. The maximum reduction in cost their approach can achieve hinges on this factor. Moreover, it also assumes that change in bandwidth due to routing additional data does not affect the overall costs.

This paper mainly involved an extensive survey of energy costs and the feasibility of such a scheme based on Akamai systems. It modeled energy consumption and the increase in routing energy to show as a proof of concept that such a routing scheme could be profitable to implement for an organization with large datacenters at different geographical locations. The effectiveness of the proposal lies in mainly latency. Offloading computations to the most energy efficient server can result in increased latency which can be an important factor for consideration for eg. to Google in comparison with the few million dollars saved otherwise.