Another HA question

gymmbo · Tue Apr 19, 2011 6:44 pm

I am finally bringing my second HA node up on line. I am quickly finding out that the targets I have created will need to be reassigned. My planned setup will be 2 Starwind HA SANs (cluster A), a 4 host servers for my VMs (cluster B), and 6 VMs running file services with each of their own targets (cluster C).

Do I have to make all my targets HA to ensure availability if SAN-A or SAN-B goes down?

Also, will there be a version where I can just take an existing target and make it HA on the fly (minus the iSCSI and disk work)? That would be the best feature saving potential headaches while keeping critical servers running.

Thanks,

Tue Apr 19, 2011 8:48 pm

You don't need to cluster HA nodes. And it's a bad idea to run them virtualized (possible but not recommended). Could you please draw down interconnection diagram of what you're trying to do? Could help quite a lot to understand your setup. Thanks!

gymmbo wrote:I am finally bringing my second HA node up on line. I am quickly finding out that the targets I have created will need to be reassigned. My planned setup will be 2 Starwind HA SANs (cluster A), a 4 host servers for my VMs (cluster B), and 6 VMs running file services with each of their own targets (cluster C).

Do I have to make all my targets HA to ensure availability if SAN-A or SAN-B goes down?

Also, will there be a version where I can just take an existing target and make it HA on the fly (minus the iSCSI and disk work)? That would be the best feature saving potential headaches while keeping critical servers running.

Thanks,

gymmbo · Wed Apr 20, 2011 9:24 pm

Maybe this description will help.

I have two physical servers that are running W2K8R2 SP1 w/ Starwind HA SAN 5.6. I have links between the servers (sync and heartbeat) and I have links to the network.

I have four physical blade servers hosting anywhere from four to eight VMs per blade. These host servers are connected to the SANs via iSCSI. I have these blades clustered so if one goes down the VMs will move to the next available host.

I have six vms (spread over the hosts) running file share which I have user profiles, My Doc redirected, and other network shares (advertised through DFS). These vms are clustered to keep the resources highly available.

I was planning on clustering the SANs but you are now telling me that I may not need to. The key is to ensure if I move resources from SAN-A to SAN-B or vice versa they will remain active.

The reason why I brought this up is because as I watch the videos the demo shows two SANs sharing a quorum target and a 20 GB target.That is all fine and dandy for demo purposes but I have a lot of VMs that I need to move from various servers to my SAN and I am working with about 4TB of space. I have/am going to have far more targets that need to be highly available. I have asked around about the best way to use Cluster Shared Volumes is to use a single target per VM versus multiple VMs in a single CSV. Because of this I am creating a lot of targets which require a lot of management versus a giant 4TB target. Sorry for the ramble.

Thu Apr 21, 2011 7:40 am

OK, I see now. You don't need to cluster SAN nodes as StarWind itself is a cluster. We do ensure disk resources and memory are synchronized between SAN nodes so if one would go down second would be the same replica of the original (failed) one. Job to re-submit I/O is done via MPIO for Hyper-V, Xen and ESX/ESXi. So now if you have configured HA with StarWind servers you have perfectly fully redundant configuration. Just make sure you have doubled switches and network wiring. Hope this helped

gymmbo wrote:Maybe this description will help.

I have two physical servers that are running W2K8R2 SP1 w/ Starwind HA SAN 5.6. I have links between the servers (sync and heartbeat) and I have links to the network.

I have four physical blade servers hosting anywhere from four to eight VMs per blade. These host servers are connected to the SANs via iSCSI. I have these blades clustered so if one goes down the VMs will move to the next available host.

I have six vms (spread over the hosts) running file share which I have user profiles, My Doc redirected, and other network shares (advertised through DFS). These vms are clustered to keep the resources highly available.

I was planning on clustering the SANs but you are now telling me that I may not need to. The key is to ensure if I move resources from SAN-A to SAN-B or vice versa they will remain active.

The reason why I brought this up is because as I watch the videos the demo shows two SANs sharing a quorum target and a 20 GB target.That is all fine and dandy for demo purposes but I have a lot of VMs that I need to move from various servers to my SAN and I am working with about 4TB of space. I have/am going to have far more targets that need to be highly available. I have asked around about the best way to use Cluster Shared Volumes is to use a single target per VM versus multiple VMs in a single CSV. Because of this I am creating a lot of targets which require a lot of management versus a giant 4TB target. Sorry for the ramble.

gymmbo · Thu Apr 21, 2011 4:18 pm

This has. One last question regarding this. Do I have to create HA targets for all of my VMs or will the HA quorum work? I wish there was a way to take existing targets and make them HA on the fly.

Thanks again.

Thu Apr 21, 2011 5:00 pm

You need to create HA for all the VMs (no need to dedicate HA per VM, 10-15 VMs/HA drive is ok)
On-the-fly migration will be available soon, in the meantime - 5 minutes to go from non-HA to HA is not a big sacrifice IMHO