CrossPerformancePowerPrediction

Physical Systems Performance and Power Dataset: We have selected 4 physical systems for our cross performance power prediction experiments. We first have collected hardware feature values from each of the four systems using the dmidecode utility. We have modified each of the 6 benchmark applications' source code to integrate PAPI-API to collect power consumption.

PAPI-API uses Intel's RAPL to read machine status registers (MSRs) to collect power consumption. We have executed modified versions of each benchmark application on each physical system to collect runtime and power consumption from benchmark logs. The hardware features, performance (Runtime) and Power are available in files /benchmark_algo/%application%/%application%_power_lab.csv e.g. for mser it is available at benchmark_algo/mser/mser_power_lab.csv, etc.

Performance and Power Dataset from Gem5 Systems with Hardware Features of Physical Systems: For validation purposes, we have to build Gem5 systems with identical hardware feature values as the physical systems. We have collected performance power for each benchmark application by executing each of these systems. The files /benchmark_algo/%application%/Gem5LabSystemConfigs.csv under each application folder consists of performance (runtime) and power for the Gem5 systems with features of physical systems.