This is an archived copy of the Xen.org mailing list, which we have preserved to ensure that existing links to archives are not broken. The live archive, which contains the latest emails, can be found at http://lists.xen.org/
Home Products Support Community News


Re: [Xen-users] Xen system hang or freeze


Two points:

1. Your domU was already hosed at 12:55pm with the 100% steal. This means that the VM was ready for CPU and the hypervisor scheduler didn't allocate a pcpu. It's worth configuring an alert to fire when %steal gets high

2. How can we explain that, when the host isn't busy, the scheduler isn't giving the vcpu some free CPU time?

I thought that, unless a domain has been configured to have additional vcpus,
it can't be scheduled on more than one pcpu.

So if, at most, three pcpus (cores) are being used by DomUs, and the Dom0 is idle, what is happening on the other CPUs?
mpstat -p ALL would help a little


On Apr 21, 2009, at 2:42 PM, Nick Anderson wrote:

On Tue, Apr 21, 2009 at 12:31:04PM -0400, Peter Booth wrote:
Some thoughts:
0. Do you have the default behavior where the guests independent
wallclocks are disabled?
Yes I did not change anything relating to that.

1. I have observed visible performance differences from a VM when %steal
goes above 1%.
It sounds like you have 8 cores.
How many VMs do you have?
What are their weights and caps?

I do have 8 cores .... 2 quad core AMD
I have 3 domUs each with a single vcpu no pinning, weight or caps just
the single vcpu per domU.

2. The system default of collecting sar every ten minutes is pretty
unhelpful for problem diagnosis. I routinely adjust this to interval to
five seconds, which for the expense of a lot of disk space, gives a
historical dataset that is useful for forensics.

I might try taking it down to the 1 minute mark first.
Should I do this in a guest as well as the host?

Nick Anderson <nick@xxxxxxxxxxxx>

Xen-users mailing list

Xen-users mailing list