Managing performance and power consumption in a server farm

We examine the problem of managing a server farm in a way that attempts to satisfy the conflicting objectives of high performance and low power consumption. A subset of servers is designated as ‘reserve’. The reserves are powered up when the number of jobs in the system is sufficiently high, and are powered down when that number is sufficiently low. Powering up takes an interval of time during which the reserves consume power but do not serve jobs. The question of how to choose the number of reserves, and the up and down thresholds, is answered by analyzing a suitable queueing model and minimizing an appropriate cost function. Heuristics and numerical results are also presented.