Performance results

**Figure 2:** Visualization of performance with respect to message size and noise. TCP is shown in wireframe, SCTP as solid colored surface, and UDP as a transparent solid surface
$\includegraphics{overview.eps}$

**Figure 3:** 1k messages showing performance drops from 10% noise
$\includegraphics{fig1.eps}$

**Figure 4:** 4k messages showing performance drops from 10% noise
$\includegraphics{fig2.eps}$

**Figure 5:** 64k messages showing performance drops from 10% noise
$\includegraphics{fig3.eps}$

**Figure 6:** 1k messages showing performance drops from 20% noise
$\includegraphics{fig4.eps}$

**Figure 7:** 4k messages showing performance drops from 20% noise
$\includegraphics{fig5.eps}$

**Figure 8:** 64k messages showing performance drops from 20% noise
$\includegraphics{fig6.eps}$

Overall performance of UDP, TCP, and SCTP is shown in Figure 2. The X axis plots the amount of noise, ranging from 0 Kbits/sec to 2000 Kbits/sec, and the Y axis shows message size, from 1 byte to 262144 (256K) bytes. This data represents the average bandwidth for the entire 60 second iperf run. At 1k message sizes, SCTP has better average bandwidth than TCP does under the worst noise conditions tested. Figure 3 shows graphs of runs at 1K message size and 10% noise (1Mbps on 802.11b). Figure 4 shows 4k message sizes and 10% noise, and Figure 5 shows 64k message sizes with 10% noise, which is too big for a UDP packet, so no data is present for UDP. Figure 6, Figure 7, and Figure 8 show 1k, 4k, and 64k message sizes with 20% noise.. One thing to note in the messages with 20% noise run is that Iperf was unable to reliable report the bandwidth on a 1-second interval, and under heavy noise, the transfer often dropped to 0 for more than a second at a time. This shows up as the dropouts ending after a fewer number of samples than other runs.

Troy Benjegerdes 2005-02-15