<div dir="auto">I don't understand numa but I have about the same setup (2x 5540s) which I've pinned respectively and everything seems alright as far as I can help perhaps, but have you looked into hugepages if that's your main issue?</div><div class="gmail_extra"><br><div class="gmail_quote">On Jan 27, 2017 1:21 PM, "sL1pKn07 SpinFlo" <<a href="mailto:sl1pkn07@gmail.com">sl1pkn07@gmail.com</a>> wrote:<br type="attribution"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="auto">Can you share the New vcpu config? <div dir="auto"><br></div><div dir="auto">Im intereses because have, more or less, the same configuration (2x xeon x5650)</div><div dir="auto"><br></div><div dir="auto">Greetings</div></div><div class="gmail_extra"><br><div class="gmail_quote">El 27 ene. 2017 6:04 p. m., "Jan Wiele" <<a href="mailto:jan@wiele.org" target="_blank">jan@wiele.org</a>> escribió:<br type="attribution"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Am Freitag, 27. Januar 2017, 08:52:50 CET schrieb Alex Williamson:<br>
> On Fri, 27 Jan 2017 11:58:05 +0100<br>
><br>
> Jan Wiele <<a href="mailto:jan@wiele.org" target="_blank">jan@wiele.org</a>> wrote:<br>
> > Hi,<br>
> > I've two E5-2670 CPUs and want to pin a VM to a node. CPU pinning works<br>
> > fine, however memory allocation seems to happen randomly. Sometimes it is<br>
> > done on node 0, sometimes on node 1.<br>
> ><br>
> > My VM config: <a href="http://pastebin.com/raw/J6trq7gJ" rel="noreferrer" target="_blank">http://pastebin.com/raw/J6trq7<wbr>gJ</a><br>
> > I'm on Arch Linux Kernel 4.8.13-1 with libvirt 2.4.0-2<br>
> ><br>
> > Any ideas?<br>
><br>
> Looks like it should work to me. I've certainly tested this sort of<br>
> setup on RHEL and Fedora in the past and it did the right thing.<br>
<br>
> Your<br>
> vcpupin actually looks a little off though, just like on the host, the<br>
> VM is going to enumerate cores then threads, this is why your numactl<br>
> info shows node0 w/ {0-7,16-23}, node1 w/ {8-15,24-31}, ie.<br>
> {cores,threads}. By interleaving the node1 pinning as you've done, the<br>
> mapping is off in the VM. For instance, the guest will think CPU0 and<br>
> CPU6 are threads on the same core, but they're mapped to CPU8 and CPU11<br>
> on the host, which are unrelated.<br>
Thanks! I will fix that. (I think I did this because of a user, who reported<br>
slightly better performance, in the old archlinux-vfio-thread)<br>
<br>
> I know that doesn't help your<br>
> numatune issue though.<br>
<br>
Do you know which part of the system will actually allocate the memory on the<br>
specified node? Should I see some information about this in the qemu command<br>
line parameters?<br>
<br>
______________________________<wbr>_________________<br>
vfio-users mailing list<br>
<a href="mailto:vfio-users@redhat.com" target="_blank">vfio-users@redhat.com</a><br>
<a href="https://www.redhat.com/mailman/listinfo/vfio-users" rel="noreferrer" target="_blank">https://www.redhat.com/mailman<wbr>/listinfo/vfio-users</a><br>
</blockquote></div></div>
<br>______________________________<wbr>_________________<br>
vfio-users mailing list<br>
<a href="mailto:vfio-users@redhat.com">vfio-users@redhat.com</a><br>
<a href="https://www.redhat.com/mailman/listinfo/vfio-users" rel="noreferrer" target="_blank">https://www.redhat.com/<wbr>mailman/listinfo/vfio-users</a><br>
<br></blockquote></div></div>