A study of the scalability of stop-the-world garbage collectors on multicores

Lokesh Gidra (1, 2) , Gaël Thomas (1, 2) , Julien Sopena (1, 2) , Marc Shapiro (1, 2)

1 Regal - Large-Scale Distributed Systems and Applications
2 MULTICORE

Large-scale multicore architectures are problematic for garbage collection (GC). In particular, throughput-oriented stop-the-world algorithms demonstrate excellent performance with a small number of cores, but have been shown to degrade badly beyond approximately 20 cores on OpenJDK 7. This negative result raises the question whether the stop-the-world design has intrinsic limitations that would require a radically different approach. Our study suggests that the answer is no, and that there is no compelling scalability reason to discard the existing highly-optimised throughput-oriented GC code on contemporary hardware. This paper studies the default throughput-oriented garbage collector of OpenJDK 7, called Parallel Scavenge. We identify its bottlenecks, and show how to eliminate them using well-established parallel programming techniques. On the SPECjbb2005, SPECjvm2008 and DaCapo 9.12 benchmarks, the improved GC matches the performance of Parallel Scavenge at low core count, but scales well, up to 48 cores.

Mots clés: Garbage collection; NUMA; multicore

Domaines: Système d'exploitation [cs.OS]

Lokesh Gidra, Gaël Thomas, Julien Sopena, Marc Shapiro. A study of the scalability of stop-the-world garbage collectors on multicores. ASPLOS 13 - Proceedings of the eighteenth international conference on Architectural support for programming languages and operating systems, ACM, Mar 2013, Houston, United States. pp.229-240, ⟨10.1145/2451116.2451142⟩. ⟨hal-00868012⟩