Optimising the Linda in primitive: understanding tuple-space run-times

In this paper we examine tuple space systems from a distributed viewpoint. We show that current implementations are pessimistic about the timing of removal of tuples from a tuple space when an in is performed; this leads to agents having to unnecessarily block and to lowering systems performance. After providing evidence of the problem by examining distributed execution traces we then describe an implementation strategy that is highly effcient and is more optimistic about tuple removal. We discuss also the generalization of the approach to support other primitives, which have been proposed as additions to Linda, such as the collect and copy-collect primitives.