Check out the new USENIX Web site. next up previous
Next: Internet Microbenchmarks Up: Results Previous: Experiment 4: Nice with

Other results

Due to space constraints, we state two other results here, but omit detailed discussions and graphs. The full discussion appears in the extended version [49].

First, we also perform experiments with synthetically generated ON/OFF Pareto UDP foreground traffic, which is much burstier and less predictable than TCP foreground flows. We observe that Nice still causes lower interference than Reno or Vegas, but does not match router prioritization as closely. The utilization of spare capacity by Nice is also lower compared to the trace workload case. This suggests that the benefits of Nice are reduced when traffic is unpredictable. Second, we compare Nice to simple rate limited Reno flows. When the rate is tuned to approximate the spare capacity of the network, rate limiting performs well. Nice, however, outperforms rate limiting and does not require hand tuning.





Arun Venkataramani 2002-10-08