He should be using all the threads for this. Using one thread to fault a bunch of memory is a recipe for terrible performance all around.<div class="gmail_extra"><br><br><div class="gmail_quote">On Mon, Dec 17, 2012 at 9:21 AM, Barry Smith <span dir="ltr"><<a href="mailto:bsmith@mcs.anl.gov" target="_blank">bsmith@mcs.anl.gov</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><br>
Should we have a test like this in the PETSc benchmark directory?<br>
<br>
Barry<br>
<br>
<br>
Begin forwarded message:<br>
<br>
> From: Kazutomo Yoshii <<a href="mailto:kazutomo@mcs.anl.gov">kazutomo@mcs.anl.gov</a>><br>
> Subject: [Xlab] slow pagefault on KNC<br>
> Date: December 17, 2012 10:48:06 AM CST<br>
> To: "<a href="mailto:xlab@cels.anl.gov">xlab@cels.anl.gov</a>" <<a href="mailto:xlab@cels.anl.gov">xlab@cels.anl.gov</a>><br>
><br>
> Hi,<br>
><br>
> I noticed that pagefault is very slow on KNC. It takes 850 usec<br>
> while it takes ~1 usec on Xeon, so prefaulting 1GB of memory region<br>
> takes 222 sec on KNC.<br>
><br>
> This may not impact a big app runs for hours and hours, but I guess<br>
> this definitely affects short-lived processes or threads, which might<br>
> make MIC less fascinating.<br>
><br>
> This could be a hardware problem(need to check Phi), a kernel bug,<br>
> or maybe sage of SMP kernel on many-core.<br>
> If this is the last case, it would be really interesting for me.<br>
><br>
> Attached a simple page fault benchmark.<br>
><br>
> - kaz<br>
[see attached file: pftest.c]_______________________________________________<br>
> Xlab mailing list<br>
> <a href="mailto:Xlab@lists.cels.anl.gov">Xlab@lists.cels.anl.gov</a><br>
> <a href="https://lists.cels.anl.gov/mailman/listinfo/xlab" target="_blank">https://lists.cels.anl.gov/mailman/listinfo/xlab</a><br>
</blockquote></div><br></div>