Performance events or performance monitoring counters (PMCs) have been originally conceived, and widely used to aid low-level performance analysis and tuning. Nevertheless, they were opportunistically adopted for energy predictive modeling owing to lack of a precise energy measurement mechanism in processors, and to address the need of determining the energy consumption at a component-level granularity in a processor. Over the years, they have come to dominate research works in this area.
Modern hardware processors provide a large set of PMCs. Determination of the best subset of PMCs for energy predictive modeling is a non-trivial task given the fact that all the PMCs cannot be determined using a single application run. Several techniques have been devised to address this challenge. While some techniques are based on a statistical methodology, some use expert advice to pick a subset (that may not necessarily be obtained in one application run) that, in experts' opinion, are significant contributors to energy consumption. However, the existing techniques have not considered a fundamental property of predictor variables that should have been applied in the first place to remove PMCs unit for modeling energy.
We propose to address this oversight in this talk. We present a novel selection criterion for PMCs called additivity, which can be used to determine the subset of PMCs that can potentially be used for reliable energy predictive modeling. We also show that the use of non-additive PMCs in a model renders it inconsistent.
