<p>Yes we do. This was something we wanted to make sure we got right.</p>
<div class="gmail_quote">On 19 Dec 2011 20:07, "Andrew Gallatin" <<a href="mailto:gallatin@cs.duke.edu">gallatin@cs.duke.edu</a>> wrote:<br type="attribution"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
On 12/19/11 14:52, Matthew Faulkner wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
That option set the affinity for the whole netperf process. I needed to<br>
set the affinity for the threads / processes (server or clients) within<br>
netperf, i.e. have two servers but running on different cores at the<br>
same time. Also I needed it to output the results coherently.<br>
</blockquote>
<br>
You don't mention in your readme, but do you synchronize your threads<br>
to prevent over-reporting of bandwidth due to all threads running<br>
not-quite-in-parallel?<br>
<br>
In the past, when I've wanted multiple "threads" of netperf pinned to<br>
different cores, I've used wrapper scripts to just start multiple<br>
copies & parse the output. The problem of not all copies starting<br>
at the same time can be somewhat mitigated by running very long<br>
tests. But a real barrier would be nice..<br>
<br>
Drew<br>
<br>
______________________________<u></u>_________________<br>
netperf-talk mailing list<br>
<a href="mailto:netperf-talk@netperf.org" target="_blank">netperf-talk@netperf.org</a><br>
<a href="http://www.netperf.org/cgi-bin/mailman/listinfo/netperf-talk" target="_blank">http://www.netperf.org/cgi-<u></u>bin/mailman/listinfo/netperf-<u></u>talk</a><br>
</blockquote></div>