An Evaporative Approach to Handle Dynamics in Diffusive Aggregation Schemes

Distributed computing in large-size dynamic networks oftenrequires the availability at each and every node globally ag-gregated information about some overall properties of thenetwork. In this context, since traditional broadcasting so-lutions become inadequate as the number of participatingnodes increases, aggregation schemes inspired by the phys-ical/biological phenomenon of diusion have been recentlyproposed as a simple yet eective alternative to solve theproblem. However, diusive aggregation requires specicsolutions to cope with the dynamics of the network and/orof the values being aggregated. While typical solutions arebased on periodic restarts (epoch-based approaches), in thispaper, we propose an original and more autonomic solution,relying on coupling diusive aggregation schemes with theadditional bio-inspired mechanism of evaporation. While agossip-based diusive communication scheme is used to ag-gregate values over a network, gradual evaporation of valuescan be exploited to account for network and value dynamicswithout requiring periodic restarts. A comparative perfor-mance evaluation shows that the evaporative approach isable to manage the dynamism of the values sensed over thenetwork in an eective way and, in the most of the cases,it leads to more accurate aggregate estimations than epoch-based techniques.