Research
Java Garbage Collector Choice Does Not Reliably Change Energy Use; Workload Intensity Does
arXiv 2608.19520 measured processor-package energy with EnergiBridge across Serial, Parallel, and G1 collectors, three Java applications, three workload intensities, and both OpenJDK and Oracle JDK. Parallel had the lowest mean energy at 839.8 J, then Serial at 857.6 J and G1 at 969.0 J, but an RCBD ANOVA found no statistically reliable collector effect, and no significant difference between the two JDK distributions. Workload intensity was the significant driver, with energy showing only a moderate correlation to execution time (r = 0.33), so aggregate collector rankings published elsewhere should not be treated as a portable tuning recommendation.
↳ Follow the thread