Severe Overhead in Network Cause by CPU

Hi,

We are running SUSE 11 SP4 as VM on the top of XEN and we have seen a weird behavior where the network latency increase and the network throughput increase as the CPU increase in the VM, by example with CPU utilization on the VM of 2-5%, the latency is under 0.2 ms and with throughput of 1400Mbps, but as soon as the CPU start increasing the latency increase and the throughput reduce, by example at 50% CPU utilization can easily reach 4 ms and the throughput reduce to 400Mbps, which seems to be to much of an impact at a reasonable CPU utilization, of course this can get worst if the CPU increase.

We have checked and there is not signs of network bottleneck on the physical server which host the VM (with a connections of 5Gbps) or the other end (another physical server with 10Gbps)

Have you ever seen this type of problem,

Zareh

zarehv,

It appears that in the past few days you have not received a response to your
posting. That concerns us, and has triggered this automated reply.

These forums are peer-to-peer, best effort, volunteer run and that if your issue
is urgent or not getting a response, you might try one of the following options:

Be sure to read the forum FAQ about what to expect in the way of responses:
http://forums.suse.com/faq.php

If this is a reply to a duplicate posting or otherwise posted in error, please
ignore and accept our apologies and rest assured we will issue a stern reprimand
to our posting bot…

Good luck!

Your SUSE Forums Team
http://forums.suse.com

It may be useful to help us understand what the VM, and the host too for
that matter, is doing to cause this level of utilization. While there
appears from your two data points to be a correlation, that does not mean
that one is causing the other; perhaps both are symptoms of something
else. Transferring data at 1400 Mbps sounds like a decent rate of
transfer, and even 400 Mbps is faster than most of my systems do at a
sustained level. If you’re transferring files, perhaps the CPU eventually
needs to work on I/O for some reason, which would both increase the
utilization, decrease the throughput due to the latency impact. Seeing
output from ‘top’ may be useful in this case so we can see how the CPU is
being used at the various times on the VM and host.


Good luck.

If you find this post helpful and are logged into the web interface,
show your appreciation and click on the star below.

If you want to send me a private message, please let me know in the
forum as I do not use the web interface often.