Abstract Transactional Cluster Computing

A lot of sophisticated techniques and platforms have been proposed to build distributed object systems. Remote method invocation and explicit message passing on top of traditional operating systems are complex and difficult to program. As an alternative the distributed memory idea simplifies and unifies memory access, but performance drawbacks caused by expensive distributed locking mechanisms obviated this approach for cluster computing. To avoid these slowdowns more and more weak consistency models try to avoid these slowdowns by burdening the programmer with explicit synchronization. Our research group has developed a fast transaction system for a distributed heap that guarantees semantically correct access to shared memory