Clustor generated 480 runs, and took around 45-70 minutes to complete each one. It would take more than 15 days to run all simulations on a single DecAlpha. Using Clustor the jobs were done in around 3 days. This includes distribution of jobs, collection and formatting of results and re-runs if a node became unavailable. This was all done transparently by Clustor. The use of Clustor significantly reduced the execution time.

The plots below were created from the results returned by the simulation. They were generated by Matlab 5.0 and show the Fuzzy loss, delay and traffic intensity versus data and step parameters.

The following plot shows network traffic loss as step and data were varied. It shows that the traffic loss increased as the network traffic, ie. data, increased. The traffic loss did not change much as the propagation delay increased. This was expected because the fuzzy logic network controller was designed to predict propagation delays. The graphs in general show that the controller is better at controlling the effects of variations in propagation delay than in traffic intensity.

The graph below shows that the fuzzy controller was more effective in controlling the effects due to variations in propagation delay than traffic intensity. The graph shows some small bumps that are inconsistent with the rest of the trend, but this can be ignored because it consisted of a very small part of the results.

This diagram further reinforced the idea of the fuzzy controller's effect. The amount of delay in the fuzzy controller increased by a large amount as the data traffic increased. There was little variation in fuzzy delay as the propagation delay increased, ie step.

The plot below shows how much the fuzzy logic controller was utilized as data and step varied. As data traffic increased, so did the controller utilization. When the propagation delay increased, however, the fuzzy utilization stayed fairly constant. This was expected because the fuzzy controller was designed to predict propagation delay. Data intensity could be predicted to a certain degree, but if a large amount of data traveled through the network it could not help but become congested. This explains why fuzzy utilization grows sharply as data traffic increased.