Power and energy analysis and modeling of high performance computing systems using WattProf

Modeling power/energy consumption of HPC applications and analyzing its correlation with application performance characteristics can reveal interesting insights and provide useful recommendations for improving power/energy efficiency. For that purpose, fine-grained tools are needed that can synchronize power/energy measurements with application activity. In this paper we use WattProf, a new power monitoring tool that enables high frequency (multiple kilohertz) direct power measurement of hardware components in a compute node (e.g., CPU, DRAM, GPU, NIC, PCIe cards, fans, hard drives, SSD). The platform includes support for streaming the collected data to the host, storing them on the controller, and streaming via network. Software enables synchronization of the measurements with applications under test to allow fine-grained application power profiling. Using this tools, we perform application monitoring and modeling on some benchmarks and mini-applications.