Communication-Aware Local Search for Distributed Constraint Optimization

Most studies investigating models and algorithms for distributed constraint optimization problems (DCOPs) assume that messages arrive instantaneously are never lost. Specifically, local search DCOP algorithms, have been designed as synchronous (i.e., they perform in iterations which each agent exchanges with all its neighbors), despite running asynchronous environments. This is true also an anytime mechanism reports the best solution explored during run of algorithms. Thus, when assumption perfect communication relaxed, properties were established state-of-the-art may not necessarily apply. In this work, we address limitation by: (1) Proposing a Communication-Aware model (CA-DCOP) can represent scenarios different disturbances; (2) Investigating performance existing specifically Distributed Stochastic Algorithm (DSA) Maximum Gain Messages (MGM), presence message latency loss; (3) latency-aware monotonic algorithm; (4) framework reporting by non-monotonic Our empirical results demonstrate imperfect has positive effect on due to increased exploration. Furthermore, proposed allows one benefit from inherent explorative heuristics.