Reducing Energy Usage in Resource-intensive Java-based Scientific Applications via Micro-benchmark based Code Refactorings [JCR]

MATHIAS LONGO; ANA RODRIGUEZ; CRISTIAN MATEOS; ALEJANDRO ZUNINO

COMPUTER SCIENCE AND INFORMATION SYSTEMS

COMSIS CONSORTIUM

Año: 2019

1820-0214

In-silico research has grownconsiderably. Today?s scientific code involves long-runningcomputer simulations and hence powerful computing infrastructures areneeded. Traditionally, research in high-performance computing hasfocused on executing code as fast as possible, while energy has beenrecently recognized as another goal to consider. Yet, energy-drivenresearch has mostly focused on the hardware and middleware layers,but few efforts target the application level, where many energy-awareoptimizations are possible. We revisit a catalog of Java primitives commonly used in OO scientific programming, or micro-benchmarks, toidentify energy-friendly versions of the same primitive. We thenapply the micro-benchmarks to classical scientific applicationkernels and machine learning algorithms for both single-thread andmulti-thread implementations on a server. Energy usage reductions atthe micro-benchmark level are substantial, while for applicationsobtained reductions range from 3.90% to 99.18%.</div></div></body></html>