[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-users] Where is my CPU going?

Ok, one more data point, this happened yesterday.
Basically one of the process in the Windows DomU started hogging the CPU; all other domains were fine, but that specific DomU became not responsive, like it used to happen on single core machines.
I have a Xeon 1260L (4 cores 2 threads) and I assigned the DomU 8 VCPUs. I see them all in task manager, but I couldn't "use" them. After I killed the CPU hogging process (Chrome) everything started to work flawlessly.

What are the best practices? Is it ok to assign all VCPUs to a DomU? Is this a possible bug in Xen scheduler or should I take into account VCPUs in the sched-credit weights? I couldn't find any good documentation about it.
Thanks in advance.

On Thu, Jan 19, 2012 at 4:03 PM, Enzo Lombardi <enzinol@xxxxxxxxx> wrote:
Thank you all, I will report back when I find out what is going on.

On Wed, Jan 18, 2012 at 12:57 PM, Luke S. Crawford <lsc@xxxxxxxxx> wrote:
On Wed, Jan 18, 2012 at 09:12:02AM -0800, Enzo Lombardi wrote:
> Hi all,
> I am trying to diagnose a lagging issue on one of my domains.
> Right now I have 3 DomU running + Dom0 and lagging is really bad. Even Dom0
> responsiveness is really low (sometimes it takes minutes to authenticate
> over SSH).

My experience?  90% of the time when something is that slow, it's disk
I/O that is the problem, not CPU.   (as someone else pointed out
DNS problems can cause big problems with network operations, which
accounts for most of the other 10%, but it sounds like more than just
network stuff is slow)

Run 'top'  or 'sar' on the dom0 and look for IOwait.   If I'm right,
if you run top in one window while you try to ssh in from another, while
the SSH is being slow, the IOwait (I think in top, it says %wa)  will
be really high, like above 70%.

When that happens to me?  the first thing I look for is a guest that is
swapping a lot.  The next thing I look for is a failing disk.  Especially
if you use consumer-grade disk rather than 'enterprise sata' when a
disk starts failing, it slows waay down.   The 'enterprise sata' stuff
tends to fail outright before that, which helps immensely.

Xen-users mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.