Check out the new USENIX Web site. next up previous
Next: Input link scheduling Up: Experimental results Previous: Disk scheduling

Output link scheduling

In the third experiment, an increasing number of clients continuously requested the same 1.5 MB document from either of two Web sites hosted at node S. Given that requests are much smaller than replies, little processing is required per request, and the requested document fits easily in the node S's buffer cache, the bottleneck resource is S's network output link. We reserved 50% of S's output link bandwidth to the Web site of interest and measured the latter's average throughput over three minutes. During the measurements, the site of interest had ten clients and the competing site had a varying number of clients. Figure 7 shows the results, which are very similar to those of Figure 6, where the disk is the bottleneck. FreeBSD and Eclipse/BSD are equally good when there is excess output link bandwidth, but when bandwidth is scarce, Eclipse/BSD is also able to guarantee a minimum output link bandwidth allocation.

Figure 7: The site of interest gets at least its reserved fraction (50%) of the output link bandwidth.
\centerline{\epsfxsize=3.5in \epsfbox{figs/}}\end{figure}

Jose Brustoloni