Beyond scalability: Swarm intelligence affected by magnetic fields in distributed tuple spaces

Abstract

Tuple Spaces have long been recognized as a simple and elegant model for parallel and distributed computing. This is mainly because of spatial and temporal uncoupling of system components, which simplifies inter-process communication as well as component inclusion and replacement. However, Tuple Spaces have shown scalability limitations when employed in large scale and highly demanding contexts. In order to deal with this problem, “bioinspired” techniques based on swarm intelligence including SwarmLinda and Anti-Over-Clustering have been proposed. This work shows that although these approaches do improve the system scalability, they end up producing an important degradation on tuple search performance, due to poor tuple placement. By applying the concept of “virtual magnetic fields” to swarms, a novel solution called Magnetic SwarmLinda is proposed. Magnetic SwarmLinda arranges tuples in expandable clusters of clusters (called “magnetic clusters”) naturally providing load balancing among the supporting computing nodes. Simulation results show that the proposed strategy outperforms previous approaches in most scenarios.