to shorten the painful duration of the test, you can tell the chrome benchmark merely to repeat 3 times each rather than 10.<br><br><div class="gmail_quote">On Mon, Feb 25, 2013 at 3:13 PM, Dave Taht <span dir="ltr"><<a href="mailto:dave.taht@gmail.com" target="_blank">dave.taht@gmail.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div>I have been collecting statistics on the behavior of the 3.7.5-2 version of cerowrt, and perhaps y'all out there can help. I've been doing four repeatable tests.</div>
<div><br></div>I feed the attached file into the chrome web page benchmark and run it 4 times:<div>
<br></div><div>1) No other load on the system</div><div>2) while running a single up and single down netperf stream to <a href="http://icei.org" target="_blank">icei.org</a> (on the east coast)</div><div>3) then while still loaded, after turning on simple_qos.sh, set for ~85% of the rated up/down bandwidth...</div>
<div>4) then killing the load, and keeping simple_qos enabled.</div><div><br></div><div>I export each detailed (you have to select it) .csv output from that benchmark to a file.</div><div><br></div><div>I do it against a bidirectional stream - one each of<br>
<div><br></div><div>netperf -l 6000 -4 -H <a href="http://icei.org" target="_blank">icei.org</a> -t TCP_MAERTS &</div><div>netperf -l 6000 -4 -H <a href="http://icei.org" target="_blank">icei.org</a> -t TCP_STREAM &</div>
<div><br></div><div>(if you have ipv6, more power to you, kill the -4. ) </div>
<div><br></div><div>then starting the web tests 10-20 seconds later. It's interesting to watch cero's bandwidth graphs while doing this.</div><div><br></div><div>There are netperf servers setup on the east (<a href="http://icei.org" target="_blank">icei.org</a>) and west coast (<a href="http://snapon.lab.bufferbloat.net" target="_blank">snapon.lab.bufferbloat.net</a>) and a few other places. It would be sane to setup your own someplace you control (from netperf's svn compiled with --enable-demo), as it helps to have a shorter RTT to truly load up the link. You can also run the rrul test (preferably for 5 minutes or more because this test idea KILLs web page load times) to get a heavier load and more graphs. </div>
<div><br></div><div>If you are up to trying this test series, please let me know, send along the details of your setup, and the 4 csv files, and perhaps your data will show up in an upcoming paper.</div><div><br></div><div>
The chrome web page benchmark can be had at:</div><div><br></div><div><a href="https://chrome.google.com/webstore/detail/page-benchmarker/channimfdomahekjcahlbpccbgaopjll?hl=en" target="_blank">https://chrome.google.com/webstore/detail/page-benchmarker/channimfdomahekjcahlbpccbgaopjll?hl=en</a></div>
<div><br></div><div>And requires you fire off chrome with --enable-benchmarking</div><div><br></div><div>You can use the version of netperf on cerowrt to generate the load if you like, but rrul doesn't run on that. Latest version of rrul is at <a href="https://github.com/tohojo/netperf-wrapper" target="_blank">https://github.com/tohojo/netperf-wrapper</a></div>
<span class="HOEnZb"><font color="#888888">
<div><br></div>-- <br>Dave Täht<br><br>Fixing bufferbloat with cerowrt: <a href="http://www.teklibre.com/cerowrt/subscribe.html" target="_blank">http://www.teklibre.com/cerowrt/subscribe.html</a>
</font></span></div>
</blockquote></div><br><br clear="all"><div><br></div>-- <br>Dave Täht<br><br>Fixing bufferbloat with cerowrt: <a href="http://www.teklibre.com/cerowrt/subscribe.html" target="_blank">http://www.teklibre.com/cerowrt/subscribe.html</a>