On Sat, May 25, 2013 12:31, Jeff Fisher wrote:> Hi,
>
> I''ve got a machine that''s hosting a number of virtual
machines (58 + the
> dom0) and I''m having some weird issues with packet loss and high
latency
> that happens when ksoftirqd spikes in CPU usage.
>
> The machine itself is CentOS 5 x86_64 running the stock Xen packages
> (3.1.2). It''s a dual quad-core w/ hyperthreading, 48 gigs of ram,
and
> using Adaptec hardware raid w/ SAS drives.
>
> Almost every 2 minutes, I''ll lose between 2 and 4 packets (while
running
> ping) to the domUs and see a couple of packets with very high latency. I
> see the packets coming in on eth0, the bridge, and sometimes on the vif
> device but the domU itself doesn''t see these packets.
>
> Has anyone seen this type of behaviour before? I''ve tried all the
> following with little success:
>
> - increasing txqueuelen
> - changing the smp_affinity settings for our network devces
> - dedicating the first 2 cores to dom0 and letting the domUs use
> the remaining 14.
>
> If anyone has any ideas or additional things I can try, I would be most
> grateful :)
I''ve been having a similar (though not identical) issue over the past
months. Can you advise the domU OS, and what drivers are being used for
the networking (and versions).
In my case, I have a single domU per dom0, running MS Windows 2003 with
latest testing GPLPV drivers (0.11.0.404), and I see 10 seconds to 2
minutes (varies, but usually around 20 seconds) of domU network traffic
loss (not 100% but 99.9%) during that period.
I''m still working on resolving the issue, but would be interesting to
get
more information on your situation. Also, running 58 domU''s on a single
dom0 sounds rather busy... what are all those domU''s doing? I assume
they
are not very busy, but could they be spiking occassionally causing the
issue? When you get the packet loss, is it across all domU''s at the
same
time, or just random?
Regards,
Adam