On-the-fly global garbage collection based on Partly Mark-Sweep

Abstract

A new garbage collection algorithm, the Gleaner, is proposed for multicomputers with asynchronous communication networks. The Gleaner algorithm is based on Partly Mark-Sweep(PMS) algorithms, which aggressively identify cyclic garbage by investigating whether a transitive closure from given objects forms in cycles and whether the closure has no objects reachable from outside of it. The major advantage of the Gleaner algorithm is better scalability that comes from PMS; it may collect inter-space cycles of garbage by tracing only a part of universe. Additionally, the Gleaner algorithm provides the following features: i) local garbage collectors may be executed on-the-fly and ii) imitators may use asynchronous message passing, and need not confirm whether such messages have arrived at their destinations. Therefore, the Gleaner algorithm achieves better efficiency as well as scalability in a distributed multicomputer environment.