Modern commodity platforms increasingly support thread-level parallelism, which must be exploited by garbage collected applications. We describe the design and implementation of a portable mostly-concurrent mostly-copying garbage collector that exhibits scalable performance on multi-processors. We characterize its performance for heap-intensive workloads on two different multi-processor platforms, showing maximum pause times two orders of magnitude shorter than for fully stop-the-world collection at the cost of some total mutator throughput.
@inproceedings{Hosking2006ISMM,
author = {Hosking, Antony L.},
title = {Portable, Mostly-Concurrent, Mostly-Copying Garbage
Collection for Multi-Processors},
booktitle = {ACM SIGPLAN International Symposium on Memory Management},
series = {ISMM},
year = {2006},
pages = {40--51},
month = {June},
address = {Ottawa, Canada},
doi = {10.1145/1133956.1133963},
acm = {http://dl.acm.org/authorize?N93664},
gscholar = {8}
}