Joshua Perry (3/19/2008)

---

You can scale out with a cluster, but you need to think about the bigger picture. Picture a four node cluster with all the nodes active, each one with it's own instance. All four instances can fail over and run on any of the four nodes, so if you lose a node you still have all your data available, though with only 75% of the performance. Still better than losing the whole thing. The data is then partitioned across the four instances.

If you really want to get crazy, just install 64bit Windows Server 2003 Datacenter Edition, and put it on a server with 160 GB of RAM and 16 processors. Divide it up into 16 VMs running 32bit Windows Server Enterprise Edition, each with 8GB of RAM and 2 processors, leaving 32GB of RAM to the host server for managing the VMs. Set up the VMs in an sixteen node cluster. Install 32bit SQL Server Enterprise Edition on each VM, again using the sixteen node cluster. You can put up to 50 instances on each node, so you can have each VM running 2 instances, all of which can fail over to any other VM. In this configuration, you can partition the data across 32 instances and still have full failover capability. Sounds horribly expensive? Except that the only major upfront expense is the main server chassis, since you can add processors and RAM as needed. Tie it into a SAN and add disk as needed. Since you're using VMs, you can also add software licenses as needed. Now if you take advantage of the virtualization option for SQL Server licensing, you save even more on licensing - you only need one server license.

http://www.microsoft.com/sql/howtobuy/virtualization.mspx