Parameter setting and exploration of TAGS using a genetic algorithm

We consider the performance of TAGS, a multi-host job assignment policy. We use a genetic algorithm to compute the optimal parameter settings for the policy. We then explore the performance of the policy using the optimal parameters, when the job size distribution is a heavy-tailed bounded Pareto distribution with parameter alpha. We show that TAGS only operates at low inter-arrival rates. At low rates it is very efficient in comparison with other standard policies. At high rates TAGS has to be combined with other policies to achieve good performance. We also show that the performance is nearly symmetrical around the value alpha = 1, with the best performance when alpha = 1