Garbage collection strategies for multiconsistent cluster operating systems

Garbage Collection is a crucial component of memory management facilities in operating systems, application environments and within applications. Modern systems provide automatic garbage collection thus alleviating the programmer from tracking garbage manually and avoiding incorrect usage. In this paper we present different garbage collection strategies for Rainbow OS, a type safe distributed operating system for PC-clusters with multiconsistent transactional memory. In addition to known garbage collection schemes such as reference counting or reference tracking we present reverse reference tracking strategies using information about data types like object dependencies and class relations to detect unused objects in memory. Additionally we describe a possibility to detect garbage using system checkpoints stored by a checkpointing facility outside the transactional distributed memory.