Distributed constraint satisfaction via a society of hill-climbers

In this paper, we compare two modified versions of Yokoo´s (2001) distributed breakout hill-climbing algorithm (DBA), mDBA-I and mDBA-II, along with six evolutionary computations referred to as societies of hill-climbers (SoHC) on 400 randomly generated distributed constraint satisfaction problems. Each SoHC is composed of S instances of mDBA-II running in parallel, where S represents the society size. The S hill-climbers communicate with one another through a distributed list of breakout elements which represent nogoods discovered at local minima. Our results show that larger society sizes are better in terms of convergence percentage and average cycle performance. However, this better performance comes at a cost of increased constraint checks and message sizes.