I’ve been struggling with this for a bit but I think we’ve finally settled in. We’ve moved a data warehouse away from virtualization because of the lack cores and the blade architeture not having enough HBAs. We went with an Active Active Passive cluster and got the database engine installed. All users were very delighted with the massive upgrade in hardware.
The dw had not been taking off as quickly as its users would have liked because of the performance constraints on the VM. Integration services had always been installed on the same server using the default instance msdb. IS was the first challenge because we couldn’t have default instances in a cluster. IS also is not a cluster aware service so we have it installed on all three cluster nodes. To connect to IS the virtual IP for production is used so in the event of a fail-over users will just have to reconnect. The tricky part comes with the package storage. We chose filesystem and are using a mesh topology DFS replication to make sure all three nodes have their packages folders in sync.
Reporting services is also not cluster-aware. To maintain high availability we chose a VM and have targeted the cluster instance to store the RS database. This would be a standard server deployment according to BOL http://msdn.microsoft.com/en-us/library/ms157293.aspx Using a DNS pointer instead of a server name will give us the ability to scale out with a load balancer if we need to later without having to change any URLs on people. With SharePoint also making headway we may not need this instance of RS as much.
Analysis services is cluster aware. That was a pleasant surprise after getting RS all figured out. The AS databases are stored on the same LUNs as the data source databases.
If I run into any major roadblocks in this architecture i’ll
strikeout the stupidity.