Algorithms for Right-Sizing Heterogeneous Data Centers

Power consumption is a dominant and still growing cost factor in data centers. In time periods with low load, the energy can be reduced by powering down unused servers. We resort to model introduced Lin, Wierman, Andrew Thereska [23, 24] that considers centers identical machines, generalize it heterogeneous d different server types. The operating of depends on its load modeled an increasing, convex function for each type. contrast earlier work, we consider discrete setting, where number active servers must integral. Thereby, seek truly feasible solutions. For homogeneous ( = 1), both offline online problem were solved optimally [3, 4]. this paper, study general time-dependent functions. develop algorithm based work approach which achieves competitive ratio 2 + 1 ϵ any > 0. time-independent functions, 1. There lower bound shown [5], so our nearly optimal. version, give graph-based (1 ϵ)-approximation algorithm. Additionally, able handle time-variable data-center sizes.