Handling dynamics in diffusive aggregation schemes: An evaporative approach

Distributed computing in large-size dynamic networks often requires the availability at each and every node of globally aggregated information about some overall properties of the network. In this context, traditional broadcasting solutions become inadequate as the number of participating nodes increases. Therefore, aggregation schemes inspired by the physical/biological phenomenon of diffusion have been recently proposed as a simple yet effective alternative to solve the problem. However, diffusive aggregation algorithms require solutions to cope with the dynamics of the network and/or of the values being aggregated solutions, which are typically based on periodic restarts (epoch-based approaches). This paper proposes an original and autonomic solution, relying on coupling diffusive aggregation schemes with the “bio-inspired” mechanism of evaporation. While a gossip-based diffusive communication scheme is used to aggregate values over a network, gradual evaporation of values can be exploited to account for network and value dynamics without requiring periodic restarts. A comparative performance evaluation shows that the evaporative approach is able to manage the dynamism of the values and of the network structure in an effective way: in most situations it leads to more accurate aggregate estimations than epoch-based techniques.