Energy Consumption Analysis of HPC Applications Using NoSQL Database Feature of EnergyAnalyzer

Abstract

A notion of increasing the energy efficiency of HPC machines or applications has reached the global HPC community forum in recent years. This has opened up several interesting possibilities that reduces the energy consumption of applications, including an energy consumption analysis mechanism which delves into the reason behind the energy consumption bottlenecks of applications. In order to easily analyze the energy consumption of applications (from machine to machine), a need for a dedicated energy consumption analysis tool has undoubtedly enthused application developers or users. In general, when applications were analyzed for performance bottlenecks in modern HPC architectures, such as, exascale machines which have more than tens of thousands of cores, a performance analysis tool might deliver a huge performance dataset. Querying such data in a short span of time can efficiently be done using document based NoSQL database systems. This paper proposes an online-based energy consumption analysis mechanism of HPC applications using EnergyAnalyzer Performance Database (EAPerfDB), a NoSQL-based performance database feature, of EnergyAnalyzer tool. The EnergyAnalyzer tool uses semantic agents in a distributed fashion to undergo the energy consumption analysis of HPC applications. In addition, the paper explores the findings of the energy consumption analysis of High Performance Computing Challenge (HPCC) benchmarks when NoSQL-based EnergyAnalyzer tool was used at the HPCCLoud Research Laboratory of our premise.