Re: new hardware + rhel6 migration

[Ray Wang <raywang gnome org>]

Dear all,

IMHO, What I'm thinking is just like you guys did, those mission
critical services should reside in non-virtualized system, so
NFS (everyone's data is on it)
LDAP (you have to have this one for authentication)
DNS (When it is down, you can not log into any server)
are better running on physically servers.

Blog, bugzilla, wiki, build server, snowy, web, git, l10n, developer
library are also critical, but not that fundamental critical, so they
can be running in VMs, and also it's easy to backup.

It's just my two cents, please feel free to correct me where I'm wrong. :)

Cheers,

On Thu, Dec 16, 2010 at 1:48 AM, Owen Taylor <otaylor redhat com> wrote:
> On Wed, 2010-12-15 at 10:18 -0700, Christer Edwards wrote:
>> On Wed, Dec 15, 2010 at 9:46 AM, Jeff Schroeder
>> <jeffschroeder computer org> wrote:
>> > label really isn't the best place for wiki and it seems like the first
>> > that should move. Are / were you still working on this? I know ray was
>> > helping do it. Do you think you could build a RHEL6 wiki vm and then
>> > that could be moved?
>>
>> Ray was/is working on this but ran into a snag regarding a Python
>> version dependency that wasn't available until RHEL6. I told him that
>> I would re-build the VM on RHEL6, but I also ran into a snag in
>> building it. I need to talk to you (Jeff) re: cobbler to see where my
>> problem is. After this is built he's going to re-rsync the data and it
>> should be ready to go.
>>
>> > dns master and ldap master should not be virtualized. Everything else
>> > other than nfs is fair game imo.
>>
>> Am I correct in my understanding that ldap is going to go on the
>> smaller of the two new machines? Should / can dns go with it? Do we
>> want to keep them separated? Remember, we do have a secondary NS
>> server now at the Canonical DC.
>
> To me having ldap and nfs and maybe mango together on a single
> non-virtualized system makes a lot of sense.
>
>> If nfs should not be virtualized where do we suggest that goes? Do we
>> want to put it just on combobox or the new R610?
>
> Combobox was bought specifically to do NFS. The ability to host VMs was
> secondary. The NFS would be run out of the host operating system to
> avoid another layer of IO indirection.
>
>> I guess the question
>> I'm getting to now is: if we virtualize pretty much everything, are we
>> getting rid of the old out-of-warranty hardware (which is just about
>> everything)?
>
> We have 3 newish (*) machines now, all with at least 8 cores and 32GB of
> memory. And we have 5 out of warranty or no-warranty machines currently
> running. In terms of cpu and memory the old machines are way ahead. More
> so with the new machines.
>
> We don't want anything mission critical running on the old machines; and
> basically everything we do is mission critical. I think we can just
> unrack machines and save the space and the power as we get things
> migrated off of them.
>
> Maybe save one for a build slave - build slaves are mostly disk
> intensive and while combobox has nice disks, I'm not sure we want builds
> constantly churning over NFS.
>
> - Owen
>
> (*) drawable and vbox went into service at the end of 2008, and have
>    3 year warranties on them. We really need to investigate extending
>    those out to 5 years before they expire.
>
>
>



-- 
Ray Wang
  - Follow your dreams