AAU Student Projects - visit Aalborg University's student projects portal
A master's thesis from Aalborg University
Book cover


Exploring the Energy Consumption of Highly Parallel Software on Windows

Authors

; ;

Term

4. term

Publication year

2023

Submitted on

Pages

15

Abstract

I takt med at CPU’er har fået flere kerner, har software fået flere ressourcer at udnytte. Denne afhandling besvarer fire forskningsspørgsmål for at vurdere, hvor meget ekstra ydeevne de mange kerner giver, og hvordan P- og E-kerner påvirker parallel software. P- og E-kerner betyder her Performance- (kraftige) og Efficiency- (strømbesparende) kerner. Forsøgene udføres på to computere, og analysen omfatter både energiforbrug og køretid pr. kerne samt ved stigende antal kerner. De fleste forsøg køres på Windows, hvor Intels Running Average Power Limit (RAPL) ikke er tilgængelig; Linux bruges som reference. På Windows vælges det bedst præsterende måleinstrument til energimåling gennem tests af flere værktøjer på både mikrobenchmarks (små, målrettede tests) og makrobenchmarks (større, mere realistiske arbejdsbelastninger), hvor målingerne sammenlignes med en ground truth.

As CPUs have gained more cores, software has more resources to use in parallel. This thesis addresses four research questions to assess the performance gains from additional cores and the impact of P and E cores on parallel software. Here, P and E cores refer to Performance (high-power) and Efficiency (low-power) cores. Experiments are run on two computers, evaluating energy consumption and execution time per core and as the number of cores increases. Most experiments are conducted on Windows, where Intel’s Running Average Power Limit (RAPL) is unavailable; Linux serves as a reference point. On Windows, the most accurate energy-measurement instrument is selected by testing multiple tools on both microbenchmarks (small, targeted tests) and macrobenchmarks (larger, more realistic workloads) and comparing their readings against a ground truth.

[This summary has been rewritten with the help of AI based on the project's original abstract]