This paper shows how to perform concurrent and distributed automatic garbage collection of objects possessing their own thread of control. The relevance of garbage collection and active objects to distributed applications is briefly discussed and the specific model of active objects used in the paper is explained. The collector is comprised of independent local collectors, one per node, and a distributed global collector. The mutator (application), the local collectors and the global collector run concurrently. An important part of this paper is the detailed presentation of the algorithms necessary to achieve correct concurrent operation among the collectors and between the collectors and the mutator. The collector builds on previous algorithms for taking snapshots in distributed systems and for detecting termination.>

Distributed automatic garbage collection of objects possessing their own thread of control is discussed. The relevance of garbage collection and concurrent objects to distributed applications is briefly discussed, and the specific model of concurrent objects used is explained. The collector comprises a collection of independent local collectors, one per node, loosely coupled to a distributed global collector. The mutator (application), the local collectors, and the global collector run concurrently. The synchronization necessary to achieve correct and efficient concurrent operation between the collectors is presented. One interesting aspect of the distributed collector is the termination algorithm.>

The authors show how to perform real-time automatic garbage collection of objects possessing their own thread of control. Beyond its interest as a novel real-time problem, the relevance of automatic management and concurrent objects to real-time applications is briefly discussed. The specific model of concurrent objects used in the paper is explained. A definition of real-time garbage collection is given and an algorithm satisfying this definition is described. An analysis of the relationship between latency, memory overhead and system size shows that this approach is immediately feasible for low-performance real-time systems or multiprocessor real-time systems with a dedicated processor for garbage collection. Future improvements in the collector's performance are outlined.>