Automated Profile-Guided Replacement of Data Structures to Reduce Memory Allocation

Data structures are essential tools in programming languages, offering various implementations optimized for factors like performance and memory efficiency. Selecting the optimal data structure is often difficult, especially within frameworks beyond the developer’s control. This paper presents an automated approach to optimizing data structures in Java applications. Using profiling, we collect allocation-site-specific usage metrics and replace standard data structures with customized, memory-efficient versions. Integrated into GraalVM Native Image, our approach reduces memory usage by up to 13.85 % in data structure–intensive benchmarks, with average reductions of 1.63 % in standard and 2.94 % in microservice-based benchmarks. While not beneficial for all workloads, our results demonstrate that automated profiling and replacement can improve memory efficiency and lay the groundwork for further optimizations.