A self-organizing client / server allocation algorithm for applications with non-linear cost functions

We subject a previously presented client-server allocation algorithm to a revised performance criterion to account for non-linear cost functions. We identify the conditions under which the original approach is not guaranteed to outperform the benchmark approach and propose a revised cost prediction method to improve performance for those scenarios. We present empirical results to show the measured benefit of the revised approach over the benchmark as well as the original work.