Distributed Tuple Spaces in Highly Parallel Systems

The tuple space abstraction is widely recognized as a powerful and general coordination model for parallel programming. However, it is based on the abstraction of a global space and therefore difficult to be efficiently implemented in highly parallel architectures. The paper discusses the implementation issues of the tuple space abstraction on large parallel systems and proposes a new implementation model suitable for massively parallel systems: the model achieves scalability by organizing the system in a hierarchical way and by encouraging the presence of multiple tuple spaces with a constrained scope. The paper describes and analyses the model for a transputer-based implementation: the hierarchical organization makes the access time to tuples proportional to the locality of the references and, in most cases, bounded by the logarithm of the system size.