Modeling Power Usage for the SZ Lossy Compressor on HPC Systems
ACM Student Research Competition: Graduate Poster
ACM Student Research Competition: Undergraduate Poster
TimeWednesday, 18 November 20208:30am - 5pm EST
DescriptionLossy compressors are becoming more prevalent due to the increasing volumes of data produced in HPC systems. Compressors are integral to computational workflows to assist in data I/O: transporting smaller amounts of data is more time and energy efficient. By modeling power consumption of lossy compressors, we work towards improving energy efficiency. In this paper, we present a characterization of the power usage of the SZ lossy compressor over several datasets. We describe the piecewise energy consumption of SZ's internal processes, while also computing the total energy consumption of the system. Through these contributions we provide a more thorough understanding of how HPC systems evolve power in their necessary compression stage. Overall our model enables us to lower power consumption 15% with a time tradeoff of 5% by scaling CPU frequency below base clock.