Performance monitoring counters (PMCs) are of great value to monitor the status of processors and their further analysis and modeling. In this paper, we explore a novel problem called PMC integration, i.e., how to combine a group of PMCs which are collected asynchronously together. It is well known that, due to hardware constraints, the number of PMCs that can be measured concurrently is strictly limited. It means we cannot directly acquire all the phenomenon features that are related with the system performance. Clearly, this source raw data shortage is extremely frustrating to PMCs based analysis and modeling tasks, such as PMCs based power estimation. To deal with this problem, we introduce a neighboring interval power values based PMC data integration approach. Based on the activity similarity of easily collected power dissipation values, the proposed approach can automatically combine distinct categories of PMC data together and hence realize the recovery of intact raw PMC data. In addition, the significance and effectiveness of the proposed approach are experimentally verified on a common task, the PMCs based power consumption modeling.
A Dialectic Data Integration Approach for Mixed Methods Survey Validation
