WARNING - OLD ARCHIVES

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/
   
 
 
Xen 
 
Home Products Support Community News
 
   
 

xen-users

RE: [Xen-users] migrating between different hardware (xen-3.1.3)

You know, I completely forgot that you could force xens 'hand' by building a
kernel without support for higher features.  I've given that advice myself
(on this list nonetheless).. I blame it on my current (sick/feverish) state.

You do lose a considerable amount of granularity in supporting cpu features
however.

Afaik enabling support generic x86 is not a viable solution.  You have to
ensure that the processor family is set to the lowest commonly supported
feature set.  Generic x86 support only switches the feature set on boot, and
won't work live, resulting in the same crash scenarios when migrating from
new->old.

I do agree that the xen hypervisor should check feature flags while
migrating, it seems like it would an easy change and prevent many crashes,
but don't ask me to do it, I'm just an end user after all.

Sorry that I gave such a incomplete / inaccurate response, the last thing I
want to do is scare a user away from this wonderful tool.

An end user really looking forward to cpu feature filters

        Richard [STD]Ein Barr.

-----Original Message-----
From: xen-users-bounces@xxxxxxxxxxxxxxxxxxx
[mailto:xen-users-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of Stephan Seitz
Sent: February 24, 2008 9:18 AM
To: b52@xxxxxxxxx
Cc: xen-users@xxxxxxxxxxxxxxxxxxx
Subject: Re: [Xen-users] migrating between different hardware (xen-3.1.3)

I don't want to argue against your decision, but you're able to build
your domU kernel with "generic x86" support.

Anyway, we made a few assumptions:
        - heterogene clusters are harder to manage by default
          (even without xen)
        - new x86 hardware supports virtually all features of
          current one.
        - hardware upgrades (10Gbit NICs, Infiniband, ...) are
          likely to be necessary earlier than x86 changes
        - regular updates and kernel patches are always necessary,
          optimization for newer hardware can be done if support
          for older hardware has become obsolete.
        - electricity tends to get expensive

Taking this assumptions, and having in mind that a cluster does not
necessarily need to cover every and all machine, our decision is clear ;)

But to your question. I don't know what kind of automation you're using
for migration, but it's not too hard to get /proc/cpuinfo of a remote
host to check the capabilities before calling xm migrate.
For complex scenarios, some work has always been done outside xen.

Cheers,

Stephan


b52@xxxxxxxxx schrieb:
> I already thought that, but hoped there is a way to tell the domU which
> cpu flags to use. Well this will be the reason my company will not use xen
> for domain management. You will never know what kind of CPUs you will buy
> in future and sure you don't want to buy a complete new cluster because
> you need one box more..
> 
> Btw. should the relocation daemon not at least check the CPU flags of the
> remote server before migration?
> 
> And there is really no way to pin the domU to a set of CPU flags?
> 
> -- Holger
> 
>> This has been brought up a few times on the list.  The problem is the cpu
>> features that the guest is using.  When you transfer a domain from the p4
>> to
>> the a64, it supports all the flags (being a generation or two newer) that
>> the p4 does, however in reverse it is not the case.
>>
>> There was some ongoing work to create a filter for domain creation that
>> would allow the 'lowest common denominator'.  However, it seems it's
never
>> made into the codetree.  It has been listed on the xen roadmap for a long
>> time now.
>>
>> In short, this doesn't work now, and unless somebody completes the
feature
>> filter, it may not work for some time.
>>
>> You should be careful to ensure that all the hosts that you transfer to
do
>> in fact support all the cpu flags that the creating host does, otherwise
>> you
>> will run into seemingly random crashes, corruption, or hangs.
>>
>> For the moment, it is only recommendable to do domain transfer between
>> identical (or same generation same lineup) cpus.
>>
>> -----Original Message-----
>> From: xen-users-bounces@xxxxxxxxxxxxxxxxxxx
>> [mailto:xen-users-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of b52@xxxxxxxxx
>> Sent: February 23, 2008 4:13 AM
>> To: Xen-users@xxxxxxxxxxxxxxxxxxx
>> Subject: [Xen-users] migrating between different hardware (xen-3.1.3)
>>
>> Hi,
>>
>> I setup a testbed todo some migration tests for work and got a problem..
>>
>> I used two P4 2.4GHz an AMD64 4200+ (SVM) and an Intel C2D 6320 1.86GHz.
I
>> had xen 3.1.3 with 2.6.18 and exactly the same versions on all systems
>> (mirrored AoE) and one 32Bit domU
>>
>> Now if I created this domU on one of the P4s, I was able to migrate it
>> between every host.
>> If I created this domU on the AMD or Intel C2D, it was only possible to
>> migrate between these both. The domU always crashed if I tried to migrate
>> it to the P4s.
>>
>> You can find the xm info of all 4 PCs and one xm log of source(AMD) and
>> one xm log of destination(P4) in close.
>>
>> Any help is appreciated and please CC me
>> Cheers,
>> Holger
>>
>>
>> _______________________________________________
>> Xen-users mailing list
>> Xen-users@xxxxxxxxxxxxxxxxxxx
>> http://lists.xensource.com/xen-users
>>
> 
> 
> 
> _______________________________________________
> Xen-users mailing list
> Xen-users@xxxxxxxxxxxxxxxxxxx
> http://lists.xensource.com/xen-users


-- 
Stephan Seitz
Senior System Administrator

*netz-haut* e.K.
multimediale kommunikation

zweierweg 22
97074 würzburg

fon: +49 931 2876247
fax: +49 931 2876248

web: www.netz-haut.de <http://www.netz-haut.de/>

registriergericht: amtsgericht würzburg, hra 5054


_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users