Fault tolerance in distributed genetic algorithms with tree topologies

We investigate the effects of communication failures in grid-based, distributed genetic algorithms with various topologies. We evaluated the performance behavior of distributed GAs under varying levels of persistent communication failures, using the sorting network problem as a benchmark application. In this experiment, we find that distributed GA with larger population size is less affected by the lower communication failure rate. However, the effect of lower communication failure on the performance of distributed GA varies with the topologies when population size is small. For all the tree topologies we investigated, when communications failures occur extremely frequently, then a significant performance degradation is observed. However, even in these extreme cases, we show that simple retry/reroute protocols for recovering from communication failure are sufficient to recover most of the performance.