Quality Network Load Information Improves Performance of Adaptive Applications

Abstract : The Joint Task Force Reference Architecture requires a Commas Server to aid client applications in adapting to changing network loads by apprising them of current and expected loads. The current Commas Server implementation estimates the network load by sending various sized packets and reporting raw performance statistics to the client. This implementation presents three problems: (1) clients interpret the statistics autonomously, (2) statistics are inaccurate due to the instantaneous collection procedure, and (3) clients also require the state of other resources to make informed decisions concerning adaptation. Development of a new Commas Server design, which solves these problems, is needed. This thesis develops a new Commas Server design and determines, through simulation, whether providing a more accurate estimate of the load could permit users of adaptive applications to obtain better performance. Simulations were run using many different situational parameters. Both the average size of the data successfully transmitted, and whether an application met its deadline, were recorded. The results of these simulations show that clients of the existing Commas Server perform much better because they adapt, but in some cases 14% to 30% of the messages do not arrive by their deadline. However, a better design that more accurately estimates loads could deliver at least 96% of the messages on time.