We have presented a dynamic many-to-many message dissemination approach that uses direct connections between overlay nodes for minimum latency and applies message forwarding and aggregation where this is the better option. Since the optimization algorithm runs locally on each node using knowledge only about the local neighborhood, it is applicable in scenarios with highly dynamic subscriptions.
In the evaluation we made three main observations: (1) The optimization process effectively shifts the load to less utilized nodes, leading to a distinctively more uniform load distribution and significantly fewer overloaded nodes. (2) The utility function allows controlling the trade-off between link utilization
(fairness) and induced latency. (3) A high clustering coefficient of subscriptions improves the performance of the algorithm.