Lambert, Jonathan (2026) A Multifactor Empirical Framework for Explaining Power and Energy Variation in Managed Runtime Systems: The Java Runtime Environment Case. PhD thesis, National University of Ireland Maynooth.
Preview
Available under License Creative Commons Attribution Non-commercial Share Alike.
Download (19MB) | Preview
Preview
Available under License Creative Commons Attribution Non-commercial Share Alike.
Download (2MB) | Preview
Abstract
The Java Runtime Environment (JRE) and its associated Java Virtual Machine
(JVM) underpin a wide range of contemporary software systems, from embedded
platforms to enterprise and cloud computing. While decades of research have
focused on improving execution-time performance through advances in just-in-time
compilation, considerably less attention has been directed towards understanding
the factors that influence power use and energy consumption within managed
runtime systems. Furthermore, existing studies have typically examined individual
runtime components in isolation, providing limited insight into how multiple layers
of runtime configuration collectively contribute to observed variation in power use
and energy consumption.
This thesis develops and evaluates a multifactor empirical framework for in
vestigating power use and energy consumption across the OpenJDK runtime
environment. The study examines the influence of OpenJDK source version, GNU
GCC build toolchain compiler version, JVM execution mode, garbage collector
variant, workload category, and microarchitectural performance event measures
on observed system-level power use and energy consumption. Experimental mea
surements were obtained using OpenJDK versions 9–14 executing the Renaissance
benchmark suite on a Raspberry Pi 4 Model B platform, with electrical mea
surements acquired using high-resolution instrumentation and microarchitectural
performance event measures collected through hardware performance monitoring.
The empirical investigation proceeds through three complementary stages.
First, descriptive and non-parametric inferential analyses examine the indepen
dent effects of individual runtime configuration factors on power use and energy
consumption. Second, interaction analyses investigate the extent to which these
relationships vary across differing workload category levels. Finally, hierarchical regression models are developed to examine the extent to which variation in power
use and energy consumption can be accounted for by progressively incorporating runtime configuration factors, workload category, and emergent microarchitectural
performance event measures. Model adequacy is assessed through regression
diagnostics, sensitivity analyses, and bootstrap resampling procedures.
The results demonstrate that the factors influencing power use and energy
consumption differ substantially in both magnitude and structure. Workload cat
egory consistently emerges as one of the strongest determinants of both outcomes,
while execution mode exerts a particularly important influence on energy con
sumption. The analyses further demonstrate that microarchitectural performance
event measures provide additional explanatory value beyond higher-level runtime
configuration factors, although their relationships differ between interpretive and
tiered execution. Comparative modelling also reveals that predictor effects are not
invariant across execution modes, with several coefficients exhibiting substantial
changes in magnitude and, in some cases, changes in direction when analysis is
restricted to execution-mode-specific observations. In addition, the results identify
an anomalous behavioural pattern associated with one OpenJDK release that is
consistently observed across multiple analyses and is substantially moderated by
execution mode.
Collectively, these findings provide a comprehensive empirical characterisation
of the relationships between runtime configuration, workload categories, execu
tion strategy, and microarchitectural performance event measures in explaining
variation in JRE JVM power use and energy consumption. Beyond the individual
empirical findings, the thesis contributes a reproducible analytical framework for
investigating managed runtime systems, integrating descriptive statistical analysis,
interaction modelling, and hierarchical regression within a unified methodology for examining software-hardware relationships in modern virtual machine environments.
| Item Type: | Thesis (PhD) |
|---|---|
| Keywords: | Multifactor Empirical Framework; Explaining Power; Energy Variation; Managed Runtime Systems; Java Runtime; Environment Case; |
| Academic Unit: | Faculty of Science & Engineering > Computer Science |
| Item ID: | 22039 |
| Depositing User: | IR eTheses |
| Date Deposited: | 16 Nov 2026 12:31 |
| Use Licence: | This item is available under a Creative Commons Attribution Non Commercial Share Alike Licence (CC BY-NC-SA). Details of this licence are available here |
Downloads
Downloads per month over past year
Share and Export
Share and Export