Results of the dynamic energy management of the HPC system Hawk¶
Ralf Schneider (HLRS)
Abstract¶
In this presentation we will present the results of the 9-month operation of the dynamic energy management system PowerSched on the HPC system hawk. To this end, we will first briefly introduce the basic functionality of the system and then present the evaluations of the energy savings that can be achieved through different operating modes. Alongside the energy evaluations, we will also present performance evaluations at global HPC system level, which were also made possible by PowerSched and were previously not possible in this quality. Based on the presented evaluations, we will finally draw conclusions for the future holistic optimization of the energy efficiency of the overall system, consisting of HPC system, data center and surrounding heat recovery infrastructure.