Algorithmic challenges for greening IT

Modern systems must trade off traditional performance goals with energy concerns, e.g., running faster lowers delays but increasespower usage. However, while there are well-developed theories and models for discussing computation, communication, and memory demandsof algorithms/systems, a theory for discussing the energy efficiency of an algorithm/system is only developing. In this talk, I willdescribe some recent work toward this goal, which focuses on dynamic capacity provisioning in data centers. Specifically, I will describework that investigates (i) how much can be saved by dynamically "right-sizing" the data center through managing the number of activeservers and (ii) how "geographical load balancing" can be used to make efficient use of renewable sources in Internet-scale systems. In bothcontexts I will present our new algorithms, which provide significantly improved performance guarantees when compared with the"standard" approaches using Receding Horizon Control. Additionally,if time allows, I will briefly discuss the our recent progress towardthe implementation and evaluation of these algorithms in industry data centers.The talk includes joint work with Lachlan Andrew, Minghong Lin, Zhenhua Liu, Steven Low, and Eno Thereska.