Date Added: Jun 2011
Operators of data centers are faced with the challenging goal of hosting applications that meet agreed service levels, at minimal operating costs. A significant part of these costs is energy related. Successfully reaching this goal requires optimal task-to-machine assignments. This activity relies on accurate energy and performance prediction. Widespread use of multi-core, multi-processor machines complicate past prediction methods. Therefore, this paper suggests to revisit task profiling, a method based on observations of actual task execution. As a first step in this direction, this paper reviews methods for task profiling, which account for the contention present in multi-core processors.