Self-Organized Over-Clustering Avoidance in Tuple-Space Systems

When it comes to communication performance, open distributed tuple-space systems depend heavily on the proximity of tuples to processes. Researchers have proposed many approaches for storing tuples in a way that processes benefit from the organization of tuples. Although some progress has been made, most of the proposed solutions fail to address the reverse problem: if most tuples are kept near the processes, the system's robustness is affected; the over-clustering of tuples in particular nodes creates a dependence to that particular node. Hence, we have a dichotomy where it is important to organize tuples in clusters near the processes, but it is equally important to avoid over-clustering. The ideal is to have a balance where tuples are clustered but not totally concentrated in very few tuple spaces (eg. one or two). This paper presents a selforganized solution to the tuple distribution problem, in which the possibility of over-clustering is considered.