Check out the new USENIX Web site. next up previous
Next: Case Study Applications Up: Internet Microbenchmarks Previous: Methodology

Results

Figure 8: Large flow transfer performance. Each bar represents the average transfer time observed for the specified combination of primary/secondary transfers. Empty bars represent the average time for a Reno flow. Solid bars represent the average time for a Nice flow. The narrow lines depict the minimum and maximum values observed during multiple runs of each combination.
\begin{figure*}\begin{center}
\begin{tabular}{cc}
\psfig{file=figures/phoneline/...
...) London & (d) Delaware\\
\end{tabular}\end{center}\vspace{-.4cm}
\end{figure*}

Figure 8 summarizes the results of our large-transfer experiments. On each of the networks, the throughput of Nice/NULL is a significant fraction of that of Reno/NULL, suggesting that periods of spare capacity are often long enough for Nice to detect and make use of them. Second, we note that during Reno/Nice and Reno/Nice8 actions, the primary (Reno) flow achieves similar throughput to the throughput seen during the control Reno/NULL sessions. In particular, on a modem network, when Reno flows compete with a single Nice flow, they receive on average 97% of the average bandwidth they receive when there is no competing Nice flow. On a cable modem network, when Reno flows compete with eight Nice flows, they receive 97% of the bandwidth they would receive alone. Conversely, Reno/Reno and Reno/Reno8 show the expected fair sharing of bandwidth among Reno flows, which reduces the bandwidth achieved by the primary flow.

Figure 9: Large flow transfer performance over time.
\begin{figure*}\begin{center}
\begin{tabular}{ccccc}
\psfig{file=figures/cable/d...
...b) London &(c) Delaware\\
\end{tabular}\end{center}\vspace{-.3cm}
\end{figure*}

Figure 9 shows the hourly average bandwidth achieved by the primary flow for the different combinations listed above. Our hypothesis is that Nice can achieve useful amounts of throughput throughout the day, and the data appear to support this statement.




next up previous
Next: Case Study Applications Up: Internet Microbenchmarks Previous: Methodology
Arun Venkataramani 2002-10-08