Hi Rob,
Short answer, it really depends on your provider. But generally using a
single good provider makes life easier and cheaper.
I just had a client where we failed to implement any high available vm
platform due to the way they had set up their uplinks. We had to use an
api to do ip address failover but even then it was not possible to move
a subnet to a different machine. KVM/IMPI/DRAC was not possible and no
fencing devices available. Long answer you have to think about all the
things just like you would when building the cloud on your own hardware
Things you have to consider:
* What kind of storage will you use local/glusterfs, iSCSI SAN,
NFS and how will this be connected to the nodes
* How will you synchronize the storage between multiple
providers/datacenter? What kind of interconnect will you have
between providers? What will this cost me (not cheap!!!)?
* What kind of interconnect is available between nodes
* How will you manage ip failover between providers (BGP,
multicast, ?)
* What kind of connection do you have between the management
server and the ovirt nodes (ipsec, openvpn,dedicated line,
* What are my SPOF's (Single Point Of Faillure)
* What kind of fencing devices are available
* Do the different providers have the same setup or is the
hardware different (predictability of performance)?
* Will you have to share infrastructure with other clients?
* What is your maximum acceptable failover/migration time between
providers?
* I can go on for a couple of pages
Kind regards,
Jorick Astrego
Netbulae B.V.
On Thu, 2014-01-16 at 08:44 -0500, Rob Abshear wrote:
Is it possible to use leased dedicated servers as ovirt nodes? In my
mind it seems likely, but input from the community would be
appreciated. I'm envisioning a controller node running on a dedicated
server leased from a provider and multiple compute nodes running on
several different providers spread out over the net. I don't require
any migration capability for the purposes of this. I manage a large
farm of servers, and if this works like I think it would, I believe it
would speed up deployment of new servers and also provide a better
central management platform. If this is possible, what would be the
requirements, pitfalls, etc.? Thanks in advance.