Unstructured agent matchmaking: experiments in timing and fuzzy matching

We investigate distributed matchmaking within an multi-agent system in which agents communicate in a peer-to-peer fashion with a limited set of neighbors. We compare the performance of a system with synchronized time to that of systems using several different models of continuous time. We find little difference between the two, indicating that the ordering of events does not play a part in computation. We also compare a system in which matches are made deterministically between discrete task categories to one in which task matches are made non-deterministically between continuous task categories. We consider several possible matching functions and show that their support is proportional to the spread of categories tolerable. This holds for matching probabilities as low as 0.01. We further show that the matching function's 'height' relates to the speed at which the system finds matches. For instance, we show that for a triangular matching function, doubling the probability of each service matching results in about a 1.6 times speedup.