An Integrated Framework for Power-Performance Analysis of Parallel Scientific Workloads
Sergio Barrachina, M. Barreda, Sandra Catalán, Manuel F. Dolz, Germán Fabregat, Rafael Mayo, Enrique S. Quintana–Ort́ı · 2013
The path towards Exascale systems will require to energetically address power consumption of future high performance computing (HPC) workloads which, in turn, urges for a better understanding of power usage. In this paper we, present an evolved framework to trace and analyze the power and energy consumption made by parallel scientific applications. The framework includes i) a flexible and extensible design that enables easy integration of different types of power measurement devices and addition of new functionality; ii) a new module that records information on processor states related to power consumption; and iii) an improved power measurement device to monitor internal direct current (DC) power consumption. This environment is thus revealed as a powerful yet easy-to-use tool to investigate and progress on the development of energy-efficient HPC applications. Keywords-power consumption; high performance computing; performance analyzers; scientific applications;