Comparative Analysis and Evaluation of Method Layout Algorithms

Abstract

GraalVM [1, 2] is a high-performance runtime that supports ahead-of-time (AOT) compilation through its Native Image technology, enabling Java applications to be transformed into compact, standalone executables with fast startup and reduced memory usage [3]. As part of the AOT pipeline, the compiler determines the layout of methods inside the final executable. The order in which methods are placed in the .text section affects spatial locality and consequently impacts memory footprint, TLB efficiency, and the frequency of page faults. Modern compilers employ heuristics and profile-guided optimizations (PGO) to arrange methods in a way that improves locality and overall execution efficiency, making method layout an important factor in runtime performance, especially for memory-constrained or latency-sensitive workloads. This work analyzes, compares, and extends several method layout algorithms implemented in the GraalVM Native Image compiler. We evaluate heuristic-based, profile-driven, and randomized approaches across a representative benchmark suite and investigate their impact on memory footprint and startup performance [4]. To enable fine-grained evaluation, we introduce extended measurement support that reports memory usage of the .text section [4]. The results highlight the performance sensitivity to method ordering and provide a deeper understanding of how code layout influences native executable characteristics. The contributions include a clarified and simplified view of existing layout strategies, new measurement capabilities, and a comprehensive comparison that lays the groundwork for further research on code layout optimization in AOT compilers.