This talk shows how operators of Internet-scale distributed systems, such as Google, Microsoft, and Akamai can reduce electricity costs (but not necessarily energy consumption) by dynamically allocating work among
data centers in response to fluctuating energy prices. The approach applies to systems consisting of fully replicated clusters of servers installed in diverse geographical locations where energy can be purchased through spot markets.
Using historical energy prices for major energy markets in the United States, as well as usage data from Akamai’s content delivery network, we should how much can be saved now, and what might be saved in the future given server technology trends.
Joint work with Asfandyar Qureshi, Rick Weber, Hari Balakrishnan, and John Guttag.
Bruce Maggs is a professor in the Department of Computer Science at Duke University. He was instrumental in starting Akamai Technologies, where as the first Vice President for Research and Development he led all engineering efforts. Since returning to academia he has retained a part-time role at Akamai as Vice President for Research. Professor Maggs’s research focuses on how to build large distributed systems. He serves on the steering committees of the Internet Measurement Conference and the HotNets Workshop, and has served on the program committees of the STOC, SODA, SPAA, PODC, SIGCOMM, NSDI, and IMC conferences.