A new take of Java 11 GC performance: the heapothesys case.

Abstract

Benchmarks are essential to scientific research as they provide a reliable way of comparing novel approaches with the academic standard. Specifically, benchmarks are widely used in Java to evaluate new JVM versions and Garbage Collectors (GC). As new benchmark suites and collectors arrive in the industry, it is fundamental to expand our comprehension of memory management by understanding how those novel strategies work. This work studies the performance of modern garbage collectors established in the industry by using HyperAlloc, a workload of Amazon’s Heapothesys Benchmark suite that precisely predicts memory allocation behavior and facilitates comparisons between GC algorithms. The analysis provided in this paper serves as a guide on how suitable Heapothesys is to evaluate modern collectors and provides insights on their performance trade-offs.