Experiment 3: Sensitivity to parameters

In this experiment we trace the effect of the threshold and trigger fraction parameters described in Section 2.2. Figure 5 shows the document transfer latencies as a function of the threshold for the same trace as above, with $S=1$ and 16 background flows. As expected, as the threshold value increases, the interference caused by Nice increases until the protocol finally reverts to Vegas behavior as the threshold approaches 1. It is interesting to note that there is large range of threshold values yielding low interference, which suggests that its value need not be manually tuned for each network. We examine the trigger fraction in the same way, and find little change in foreground latency as we vary this fraction from 0.1 to 0.9 (graph omitted).

Figure 6: Spare capacity vs Latency with RED

Figure 7: Number of BG flows vs Latency with RED