Add to ORKGA interesting feature of some recent parallel computers is the fact that the underlying transport mechanism behind the currently dominating message passing interfaces is based on a global address space model. By accessing this global adress space directly most of the inherent delays for administering message buffers and queues can be avoided. Using this interface we have implemented a user level distributed shared memory layer using the virtual memory protection mechanisms of the operating system. The synchronisation required for maintaining the coherency of the memory is addressed by implementing a distributed shared lock which exploits the remote atomic store operations provided by the Meiko CS-2. This allows an asynchronous stype of prog...