Collector-attributable energy–performance trade-offs in individual applications

Determine whether garbage-collector selection produces an energy–performance trade-off in individual Java applications outside the three applications evaluated in the study.

Background

The study evaluates Serial, Parallel, and G1 garbage collectors across three Java applications and finds no statistically significant collector-related effect on either raw energy consumption or Energy–Delay Product in the aggregate data. Although Serial and Parallel have lower descriptive Energy–Delay Product means than G1, the observed differences are not statistically reliable under the study’s experimental design.

The authors explicitly acknowledge that the experiment cannot exclude the possibility of a collector-related energy–performance trade-off in individual applications beyond the evaluated set. Establishing whether such application-specific effects exist would require broader application coverage and measurements capable of resolving differences that may be obscured by heterogeneous workloads and substantial measurement variability.

References

The experiment therefore did not identify a collector-attributable energy--performance trade-off, though it also cannot rule one out in individual applications outside the evaluated set.

— Garbage Collection and Energy Consumption in Java: A Controlled Study Across Workloads and JDKs  (2608.19520 - Sharma, 20 Aug 2026) in Section 6, Discussion, subsection “Interpretation of the Findings,” paragraph “Energy and runtime were positively but only moderately associated.”