Asynchronous complete distributed garbage collection

Most distributed garbage collection (DGC) algorithms are not complete as they fail to reclaim distributed cycles of garbage. Those that achieve such a level of completeness are very costly as they require either some kind of synchronization or consensus between processes. Others use mechanisms such as backtracking, global counters, a central server, distributed tracing phases, and/or impose additional load and restrictions on local garbage collection. All these approaches hinder scalability and/or performance significantly. We propose a solution to this problem, i.e., we describe a DGC algorithm capable of reclaiming distributed cycles of garbage asynchronously and efficiently. Our algorithm does not require any particular coordination between processes and it tolerates message loss. We have implemented the algorithm both on Rotor (a free source version of Microsoft .Net) and on OBIWAN (a platform supporting mobile agents, object replication and remote invocation); we observed that applications are not disrupted.