On Fri, 22 Jun 2007 03:40:48 +0100, "Daniel P. Berrange" <berrange@xxxxxxxxxx>
wrote:
>On Thu, Jun 21, 2007 at 08:33:51PM -0400, Tim Boyer wrote:
>> On Thu, 21 Jun 2007 09:09:39 -0500, Richard Blocker <rblocker@xxxxxxxxxxxx>
>> wrote:
>>
>> >I'm new to running Xen, so maybe I missed something, but this is a
>> >puzzling problem that I didn't find on any of the Xen lists.
>> >
>> >I'm running the Xen 3.0.3 that comes with RedHat EL 5 (all stock, all up
>> >to date as of today) and I have a single guest domain HVM also running
>> >RHEL 5. Whenever the guest OS runs the initial prelink job defined in
>> >/etc/cron.daily (/usr/sbin/prelink -av -mR -f) the guest CPU pegs at
>> >100% and the system stops responding. It never resumes (at least not for
>> >12 hours). I can reboot the guest domain from the host machine, and it
>> >recovers fine, until prelink runs. I even ran cpu-burnin on the HVM to
>> >see if it was just the load, but it was fine while that ran. If I run
>> >the prelink command manually, it immediately freezes.
>> >
>> >For the record, the hardware is a dual quad core Xeon system with 8GB of
>> >memory. The guest HVM uses a single CPU with 512MB of memory allocated.
>> >
>> >Has anyone else seen this?
>>
>> Hah! Someone else with this...
>>
>> I've got a trouble report into RH on this one. It's not just prelink; I
>> think
>> it's tied to rpm. I can bring the system down doing a rpm -Va, or a
>> sysreport
>> without the -norpm switch. Also a dual quad core, with 16gb, and 1gb on the
>> guest.
>
>Actually it is prelink - rpm -Va will call out the prelink libraries when
>verifying IIRC. Anyway, this is ultimately a hypervisor bug in Xen 3.0.3
>which should be fixed in the Xen 3.1.0 hypervisor. So should be working
>come RHEL-5.1 updates.
>
>Seriously though, you really really really don't want to run any OS in
>full-virt if its capable of paravirt. You'll get x10 -> x100 the I/O
>performance if you use paravirt and be able to scale up the number of
>guests per host much better. So I'd recommend using RHEL-5 paravirt
>at which point you won't see the HVM bug anymore either...
>
>Dan.
Dan, thanks _very_ much for the info. Maybe you can answer my other open
ticket while we're here. :)
I suspected that I made a Bad Choice when I did full virt on the RHEL5 systems
instead of paravirt. I've got five systems running full - is there any good
way to migrate a system from full virt to paravirt?
--
tim boyer
tim@xxxxxxxxxxxxxx
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users
|
Home