Australian Computer Society
In this paper, the authors present an algorithm which may be used to efficiently search for and retrieve tuples in a distributed tuple space. The algorithm, a core part of the Tupleware system, is based on the success or failure of previous tuple requests to remote nodes in the system, and this data is used determine the relative probability of particular remote nodes being able to fulfill subsequent future requests. The logic of this algorithm is distributed and decentralized: each node dynamically calculates its relationship with other nodes at runtime. The behavior of the algorithm using two applications is analyzed, and shows significant improvement in terms of efficiency and performance compared to comparable tuple space implementations.