[Users] new oVirt installation recommendations

[Jason Keltz]

On 10/07/2013 02:35 PM, Itamar Heim wrote:
> On 10/07/2013 06:13 PM, Jason Keltz wrote:
>> I've been experimenting with oVirt 3.2 on some old hardware, and am now
>> preparing to buy new hardware for using oVirt 3.3 in production.  I'm
>> interested in any feedback about what I plan to purchase.  I want to
>> keep the setup as simple as possible.  Our current environment consists
>> of mostly CentOS 6.4 systems.
>>
>> The combined oVirt engine and file server will be a Dell  R720 with dual
>> Xeon E5-2660 and 64 GB of memory.   The server would have an LSI 9207-8i
>> HBA connected to the SAS backplane.    The R720 enclosure has 16 x 2.5"
>> disk slots.  I would get 2 x 500 GB NLSAS drives for mirrored md rood
>> (raid1), use 12 slots for RAID10 SAS 10K rpm drives (either 600 GB or
>> 900 GB), and have an additional 2 spares.   Data storage would be
>> virtual machines, and some associated data.  The O/S would be CentOS 
>> 6.4.
>>
>> The nodes would be 3 x Dell R620, dual Xeon E5-2690, 128 GB memory, each
>> with just a single, small NL SAS root drive.  There would be no other
>> local storage.  All VMs would use the file server as the datastore.  The
>> nodes would run oVirt node.
>>
>> In terms of networking, each machine would have 4 ports - 2 x 1 Gb
>> (bonded) giving machines access to "public" network (that we do not
>> control).  The 2 x 10 Gb copper would be connected to a locally
>> installed copper 10G switch that we fully control - 1 port used for
>> storage, and  1 for management/consoles/VM migration.
>>
>> A few additional notes ...
>>
>> I chose to stick with software raid MD on the file server, mostly for
>> cost, and simplicity.  I have a lot of experience good with MD, and
>> performance seems reasonable.
>>
>> I would have gone SSD for the file server root disk, but the cost from
>> Dell for their SSD is prohibitive, and I want the whole system to be
>> included in the warranty.  NLSAS is the cheapest disk that will have
>> support for the duration of the warranty period (with Dell servers, SATA
>> drives are only warranted for 1  year).
>>
>> The nodes with 1 NLSAS drive... I've thought about replacing that with
>> simply an SD card.  It's not clear if this the best solution, or how
>> much space I would need on that card.  At least when I configure via the
>> Dell web site, the biggest SD card it seems I can purchase with a server
>> is 2 GB which doesn't seem like very much! I guess people guy bigger
>> cards separately.   I know a disk will work, and give me more than
>> enough space and no hassle.
>>
>> I've chosen to keep the setup simple by using NFS on the file server,
>> but I see a whole lot of people here experimenting with the new Gluster
>> capabilities in oVirt 3.3.  It's not clear if that's being used in
>> production, or how reliable that would be.  I really can't find
>> information on performance tests, etc with Gluster and oVirt, in
>> particular, with comparison of NFS and Gluster.   Would there be a
>
> gluster is still not available for centos 6.4, and there are some 
> issues with snapshots around it still for libgfapi.
> for posixfs, its supported since 3.2.
>
Ok. I guess it's probably best that I stick with NFS for this time around.

>> performance advantage to using Gluster here? How would it work? by
>> adding disk to the nodes, and getting rid of the file server (or at
>> least turning the file server into a smaller engine only server)?  How
>> would this impact the nodes in terms of their ability to handle VMs?
>> (performance?)  I presently have no experience with Gluster whatsoever,
>> though I'm certainly never against learning something new, especially
>> should it benefit my project.  Unfortunately, as I'm positive everyone
>> can attest for is that it's just trouble finding the number of hours in
>> the day :)  There's one thing for sure - Gluster itself, while maybe not
>> TOO complicated is still more complicated than an NFS only setup.
>
> I don't have details on this, and hope others have.
> but you are correct its an entirely different deployment architecture 
> between a central nfs server, and distributed storage on the nodes.
>
It would be helpful if the documentation for oVirt had more information 
on this.

>>
>> As I've mentioned before, we don't use LDAP for authentication, so I'll
>> be restricted to one admin user at the moment unless I setup a separate
>> infrastructure for oVirt authentication. That will be fine for a little
>> while.  I understand that work may be underway for pluggable
>> authentication with oVirt.  I'm not sure if that ties into any of the
>> items on Itamar's list though. Itamar? :)  I was hoping to see that
>> pluggable authentication model sooner rather than later so that I could
>> write something to work with our custom auth system.
>
> well, you could also launch an openldap/ipa/ad/etc. in a VM. of course 
> if it has issues you'd need admin at internal to fix it.
>
I was thinking of doing this if I had to, but it's still a lot of 
headache for a few logins.
Is the pluggable authentication coming in a new version of oVirt?
>>
>> In terms of power management - my existing machines are using a Raritan
>> KVM with Raritan power management dongles and power bars. I haven't had
>> an opportunity to see if oVirt can manage the devices, but I guess if
>> oVirt can't do it, I can continue to manage power through the KVM
>> interface.
>
> are they supported by fence-agents in centos?
>
I've never tried.  I don't often need to power off hosts the hard way.. 
a reboot it usually fine. When I do need to power manage hosts, I go 
into the Raritan KVM, click on the host, and turn it off, and back on, 
and everything is fine.  I haven't done any connection to Linux.

>>
>> Any feedback would be much appreciated.
>>
With your experience with oVirt, any feedback on the hardware/NFS server 
combination?

Jason.