Benchmark data: Optimising weak reference processing in the jvm z garbage collector.

Benchmark measurement data collected for the Master's thesis "Optimising Weak Reference Processing in the JVM Z Garbage Collector" (Uppsala University, 2026). The dataset contains GC timing statistics and memory usage samples from two custom microbenchmarks designed to evaluate four optimisations to the weak-reference processing pipeline of Generational ZGC in OpenJDK 26. Nine optimisation variants are compared: the unmodified baseline (none), seven combinations of three WeakReference pipeline optimisations (optimised clear path, skip-enqueue for queue-less references, and dynamic discovered-list arrays), and a weak_fields variant that replaces WeakReference wrapper objects with a prototype @weak field annotation processed directly by the GC. Benchmarks Single-object: holder objects each holding a WeakReference (or @weak field) to a single shared target; all references become clearable in one GC cycle, isolating pipeline throughput. Multi-object: holder objects with randomly sized byte-array payloads; strong references are nulled in five rounds of 20 % each, spreading reference processing across multiple GC cycles. Execution environments UPPMAX Pelle cluster (runs slurm-5076215 and slurm-5118807): AMD EPYC 9454P (Zen 4), 48 physical cores, 96 threads, 768 GiB RAM, Rocky Linux 9.7. UPPMAX is part of NAISS, partially funded by the Swedish Research Council through grant agreement no. 2022-06725. Development laptop (run local-20260529-160736): 12th Gen Intel Core i7-12700H (6 performance cores, 8 efficient cores), 16 GiB RAM, Ubuntu 24.04.4 LTS. Files All measurement files follow the naming convention <run-prefix>_combined_<type>.csv. The run_prefix column in each CSV file identifies the originating run, and full run configuration and hardware metadata for every run are recorded in configs.json. slurm-5118807_combined_gc.csv — primary GC statistics dataset. 250 measurement iterations (plus 1 warm-up), 2 million objects (multi-object) / 20 million (single-object), 4 parallel JVM instances, 100 GiB heap. slurm-5118807_combined_memory.csv — high-frequency memory samples (RSS, GC committed memory, Java heap) for the slurm-5118807 run. slurm-5076215_combined_gc.csv — earlier GC statistics dataset from the same cluster. 100 measurement iterations, 1 million objects (multi-object) / 10 million (single-object), 4 parallel JVM instances, 100 GiB heap. slurm-5076215_combined_memory.csv — memory samples for the slurm-5076215 run. local-20260529-160736_combined_gc.csv — small-scale GC statistics run on the development laptop. 50 measurement iterations, 500 thousand objects (multi-object) / 5 million (single-object), 1 parallel instance, 8 GiB heap. local-20260529-160736_combined_memory.csv — memory samples for the local run. configs.json — full run configuration and hardware metadata for all three runs.