Simulation-based performance analysis of channel-based coordination models

In this study we consider a software system where subsequent processing steps are served by a common pool of servers, each of which can be assigned to the jobs pending at each one of the processing steps. For this model, we study the problem of finding the dynamic server assignment policy that minimizes the average long-term cost per time unit. For the two-node case, we derive a full characterization of the optimal server assignment policy. Moreover, numerical analysis reveals the highly remarkable observation that truncation of the state space has a dramatic impact on the calculations of the optimal policy for systems with three or more nodes. The results also show that this complication can be effectively overcome by a smart smoothing algorithm.