I’m in the process of developing a presentation entitled “Building Your First SQL Server Cluster” for some upcoming SQL Saturdays, and some inspiration for a few blog posts have come out of it. This one is short and fairly simple.
Windows, as we all know and love, has us use drive letters as the root of file systems. As there are only 26 letters in our alphabet, combined with the fact that A, B, and C generally aren’t available, leaves us with only 23 or 22 available for a given cluster. While this may seem like a lot, if you are splitting out tempdb, transaction logs and backups, you can chew up a lot of letters in a hurry. If you are using multiple instances in your cluster, the problem is compounded. And remember–the cluster can only see drives that are attached to it–you can’t restore a backup from a UNC path.
So what is the solution to this conundrum? Mount points–mount points allow you to mount multiple physical devices under one drive letters, where they look similar to directories. This does prevent a bit of a monitoring challenge, but that’s best left for another post.
I’ve posted before on how to add a disk to a SQL Server cluster here. Using mount points is really straightforward–assuming you have 3 “disks” (could be SAN luns or Windows volumes (well not in a cluster), designate one of the drives to be the root (the drive letter S:\ in this case).
*Note–it doesn’t really matter which disk you choose to be the root–in my case I select the backup lun, but I discussed this with a couple of others, and no one knows of any impact to system performance.
When you are assigning a drive letter to the next device, stop–select the option to “add to empty NTFS folder”, make sure you have the root drive highlighted (Windows likes to default to C:\) and name your folder (you can easily rename through explorer later).
One note–when you go to add the disks to your cluster (if newly creating), you will only see the root disk as available, the other drives won’t be selectable. This is expected behavior and fine. Good luck and happy clustering.