Heap-Snapshot Matching and Ordering using CAHPs - A Context-Augmented Heap-Path Representation for Exact and Partial Path Matching using Prefix Trees

Watch this 14-minute conference presentation from OOPSLA 2025 that introduces a novel approach to optimize GraalVM Native Image startup performance through improved heap-snapshot matching and ordering. Learn about Context-Augmented Heap Paths (CAHPs), a new representation method that enables precise matching of semantically equivalent objects across different Native Image binaries. Discover how this technique addresses the critical challenge of object identification in heap snapshots, where objects lack unique identities and heap contents vary between builds due to nondeterminism in the image-build process. Explore the implementation of prefix trees for exact and partial path matching, and understand how profile-guided optimization can be enhanced through better object mapping between instrumented and optimized binaries. Examine the experimental results demonstrating significant improvements in startup performance, including a 2.98× reduction in page faults and 1.98× faster startup times compared to the original Native Image implementation. Gain insights into the technical details of heap-snapshot reordering strategies and their impact on Function-as-a-Service and Serverless workloads that rely on Ahead-of-Time compilation for performance optimization.