Multi-threaded design for a software distributed shared memory system

This paper describes the design and implementation of a multi-threaded Dist ributed Shared Memory (DSM) system, called Cohesion, which provides high pr ogramming flexibility and latency masking, and supports load balancing. Coh esion offers a parallel programming environment which is very similar to th at on a multiprocessors system. Threads could be created recursively in thi s environment, and users are not required to handle the locations of the th reads. Instead of supporting a shared variable model, Cohesion provides a g lobal shared address space among all nodes in the system. The space is furt her divided into three regions, i.e., release, conventional, and object-bas ed memory, each is applied with different consistency protocol. In this pap er, the design issues in an ordinary thread system, such as thread manageme nt, load balancing, and synchronization, have been reconsidered with the me mory management provided by the DSM system. Several real applications have been used to evaluate the performance of the system. The results show that multi-threading usually has better performance than single-threading becaus e the network latency can be masked by overlapping communication and comput ation. However, the gain depends on program behavior and the number of thre ads executed on each node in the system.