[ovirt-users] ovirt high points

[Christopher Cox]

On 09/07/2017 06:33 AM, david caughey wrote:
> Hi Folks,
> 
> I'm giving a demo of our new 3 node oVirt deployment next week and am 
> looking for some high points that I can give to the Managers that will 
> be a sell point.

Could be hard to sell.  It's not like VMware (all in all) is deficient 
functionality wise.

> If you could help with the below questions I would really appreciate it:
> 
> Who are the big users of oVirt??

We use oVirt in production.  We have about 130 VMs on a 9 node cluster 
using Dell blades.  It houses both our test and production VMs.  We have 
a separate oVirt setup for development hosting probably about 20 VMs 
(maybe less), it's an 7 node cluster (but much lesser blades there).

In both cases they are connected to Equalogic iSCSI SAN equipment with 
multiple tiers of storage.  Each production blade has 4 x 10gbit iSCSI 
(multi)paths to storage.  The production blade subsystem uses multiple 
40Gbit links, for iSCSI storage and for LAN. Just 10Gbit links and 1Gbit 
paths on the development blades and subsystem.

Both use a dedicated oVirt management host.

The production(and test) blades run oVirt 3.6 and the dev blades are 
oVirt 3.5.

About 2 years ago we migrated our production blades from oVirt 3.4 on 
older blades and older SAN equipment to oVirt 3.6 on new blades and new 
SAN storage.  We used oVirt's export domain to facilitate the move.

We will be migrating off the development cluster and we are setting up a 
new cluster on the same DC as our production area which will be used to 
house both test and development.  Thus we are moving to just the one 
oVirt 3.6 (we're adding 5 extra blades for that cluster).

Btw, our VMs include multiple version of CentOS, Windows Server and 
Windows desktops (and even some docker nodes, but we're redoing all of 
that).  Our VMs include about 10 large PostgreSQL database servers, some 
MySQL, several Jboss servers, many web microservices (Springboot) 
servers and lost of application infrastructure servers.

> 
> Why oVirt and not vMware??
> (we are a big vMware house so free doesn't cover it)

Uh free, and to be honest, that's the best reason to do this IMHO.

> 
> What is the future for oVirt??

Unknown.  But pretty sure Red Hat will want to keep RHEV around, which 
means oVirt probably will be here for quite some time.

> 
> Why do you use oVirt??

Free.

> 
> Any links or ideas appreciated,

oVirt is NOT VMware.  But if you do things "well" oVirt works quite 
well.  Follow the list to see folks that didn't necessarily do things 
"well" (sad, but true).

I inherited this oVirt... not ideal for blades because it's better to 
have lots of networks.  We just have two blade fabrics, one for SAN and 
one for the rest, and it would be nice to have ovirtmgmt and migration 
networks be isolated.  With that said, with our massively VLAN'd setup, 
it does work and has been very reliable.  For performance reasons, I 
recommend that you attempt to dedicate a host for SPM, or at least keep 
the number of VMs deployed there to a minimum.  There are tweaks in the 
setup to keep VMs off the SPM node (talking mainly if you have a 
massively combined network like I have currently).

We've survived many bad events with regards to SAN and power, which is a 
tribute to oVirt's reliability.  However, you can shoot yourself in the 
foot very easily with oVirt... so just be careful.

Is VMware better?  Yes.  Is it more flexible than oVirt?  Yes. Is it 
more reliable than oVirt? Yes.  In other words, if money is of no 
concern, VMware and VCenter.

We will likely never do VMware here due to cost (noting, that the cost 
is in VCenter, and IMHO, it's not horrible, but I do not control the 
wallet here, and we tend to prefer FOSS here... and FOSS is my personal 
preference as well).

Companies generally speaking just want something that works.  And oVirt 
does work.  But if money is of no concern and you need the friendliness 
of something VCenter like (noting that not everyone needs VCenter or 
RHEV-M or oVirt Manager), then VMware is still better.

If you don't need something VCenter like, I can also so say that libvirt 
(KVM) and virt-manager is also reasonable, and we use that as well.  But 
we also have a (free) ESXi (because we have to, forced requirement).

The ovirtmgmt web ui is gross IMHO.  It's a perfect example of an 
overweight UI where a simplified UI would have been cleaner, faster and 
better.  Just because you know how to write thousands of lines of 
javascript doesn't mean you should.  Not everything needs to act like a 
trading floor application or facebook.  The art of efficient UI design 
has been lost.  With that said, the RESTful i/f part is nice.  Nice to 
the point of not needing the SDK.

Finally, VMware can be expensive.  It's not a "one time" purchase.  It's 
HAS TO BE ongoing.  And it can get very expensive if not understood. 
With that said, if you have anything Microsoft in the enterprise, you 
already understand and are prepared to throw cash for IT infrastructure. 
  If you do go VMware, make sure to use a hefty Vcenter host as upgrades 
to VCenter involve a lot of bloat and waste.

VMware can be a real "pain" support wise.  They can deprecate your 
entire hypervisor HW stack, especially true in a major release.  They 
can even deprecate HW in a minor release (I have fallen victim to this).

Thus, again, if you have money to burn and have relatively short HW life 
cycles (less than 5 years for sure), AND that includes OS life cycles as 
well, then VMware is probably ok.  Not saying there aren't some problems 
on the oVirt side as well, just saying VMware has more expensive warts. 
And thus "paid support" becomes somewhat humorous (but in a sad sort of 
way).

(oVirt community support ROCKS!  Just saying...)