Need guidance in setting up a SQL 2008 Failover cluster on ESXi

[macpiano]

Structure is 2 subnets connected with up to 10 gig with the San in each about 1-2 gigs on the nics. I want to setup a 2 node SQL failover cluster with a W2k8 server in each subnet but it looked like the 2 W2k8 servers needed to be failover as well. It also looked like the Servers needed shared storage which is not a problem but I would rather they don't share that. I would rather they use the 2 seperate SANs. Environment is Esxi 5 with Vcenter etc. Servers are Cisco Blade servers in each location.

thanks for any help.

[yannbizeul]

I don't have a full answer to that, but you are talking about not sharing the data... So you need to check software cluster solutions that would handle replication. Or maybe consider SRM with array or host based replication.

The point is that if you don't share storage, you have to replicate in some way, and VMware is only aware of replication at the SRM level.

You can also consider third party solutions like quest's or veeam's

Or maybe you don't need replication at all, and VMware HA is what you are looking for. For further assistance maybe you can elaborate the SLA you're targeting

[Spiffman192]

It depends on the type of failover cluster your trying to acheive.  If you're trying to utilize an MS cluster, you'll need at least a quorum and data shared store for the cluster.  SQL will sit on top of the clustered services and failover as needed or desired.

To do that within VMware, you'll need to create RDMs that are shared by the VMs.  Follow this document:

Settup for Failover Clustering and Microsoft Cluster Service

Apart from that, third party solutions will require a replication process and will exist largely outside of ESXi / VMware control.

Hope that helps.

[macpiano]

I think we would be happy with a SQL server in one subnet that replicates its database to another sql server. We are not in an environment where we need an immediate failover if the main SQL server gets hosed.  We just want to be able to get back and running without having to build a new server and do a restore. Can't use Fault tolerance because you can only use 1 cpu on the server and well a database is one you would need more than 1 cpu.

[yannbizeul]

Then use VMware HA. It provides same protection level as FT, just adding the reboot time.

be aware that neither FT or HA provides a way to change the IP address of the server, you'll need do it manually if needed, or provide the IP through a DHCP server with static IP allocation

[macpiano]

Our 5 servers in the Cisco UCS are already in HA and they keep the same mac and ip address when they move. Or is there a way to span across to the other UCS host for HA? I'm fairly new to this whole part of it.

[yannbizeul]

No you're right, my idea was stupid since a VM always keeps its MAC address

But why would you need to restart the server on another chassis ? What type of failure should be handled ?

[macpiano]

Like I said I'm just geting into this.

[yannbizeul]

Asked otherwise, what would be wrong with simple HA :

In case the host fails, all of its VM will be restarted on another host and the service can continue. You can even configure HA to detect failure at the OS level : if the host do not receive a signal sent by VMware Tools in the OS after a certain delay, it will reboot the VM, and the service comes back.

Now if you are talking about BRP and willing to get back into business after a datacenter failure for example, and you need to start your VM on another physical location with a different IP plan, this is another story.

[Spiffman192]

Not to muddle the conversion, but if there are apps dependant on this SQL, the HA recovery time-period may cause problems.  Eyes wide open going in is important.  Look at all the dependancies, decide if some type of more aggressive availability may be in order.  If you have a physical SQL cluster currently, it may still need to exist on the virtual side as well.