Solving DCSP problems in highly degraded communication environments

Although there have been tremendous gains in network communication reliability, many real world applications of distributed systems still face message loss, limitations, delay, and corruption. Yet despite this fact, most Distributed Constraint Satisfaction (DCSP) protocols assume that communication is perfect (messages that are sent will be received) although not ideal (not in a timely manner). As a result, many protocols are designed to exploit this assumption and are severely impacted when applied to real world conditions. This study compares the performance of several leading DCSP protocols including the Distributed Stochastic Algorithm (DSA), Distributed Breakout Algorithm (DBA), Max-Gain Message (MGM) and Distributed Probabilistic Protocol (DPP) to analyse their behaviour in communication degraded environments. The analysis begins by comparing the performance of all of the protocols in a perfect communication environment. We then use a simulated communication degraded environment where messages are probabilistically lost. Finally, we compare their performance by limiting the communication rate, which introduces delay. We show that DBA, once modified with a message timeout, is quite resistant to high message loss while DPP and DSA converge slower onto worse solutions. Our results also show that the setting of timeout value for DBA and MGM is an important factor in the convergence of these algorithms. Under conditions of message delay, DPP and DSA are less affected than DBA and MGM. Overall, DPP and DSA cause considerably less network load.