Hi NavGee,
I spend a fair amount of time working with NetApp products/technologies.
SnapMirror is a technology that synchronizes two volumes (or, less commonly, two qtrees). It operates at block level. The most common purpose of using this is to synchronize a volume between two separate filers. They don't have to be "like" in terms of model, and can be one-to-one, one-to-many, one-to-one-to-one (chained, kinda like how SLPs work in passing data along), etc. You can do synchronous (blocks and nvram stay consistent), asynchronous (basically it replicates snapshots), or semi-synchronous (does its best to stay synchronous, but less penalty on the active filer).
For clarity, in principle, you don't "need" snapmirror/snapvault/etc. to back up a NetApp via NDMP. You can define a volume, put data on it, configure NetBackup appropriately to communicate with it (defining networking, storage, the NDMP credentials, a policy, etc.), and just do an NDMP dump straight from the volume to your storage. NetBackup will create a snapshot on the live volume, back that up (in the form of a level 0 dump), and delete the snap when done. However, depending on sizing, usage, timing, etc. this may put considerable load on the active/production system, and care needs to be taken.
That performance issue is what it sounds like they're attempting to address in mentioning snapmirror, as you can back up the destination volume via NDMP and avoid load on the source. In this case, you'd want to ensure "big picture" orchestration was considered in terms of ensuring that the data makes its way to the mirror-target volume at an appropriate time prior to backing it up to ensure RPO is met, but other than that it's not all that relevant that you're mirroring. Just NDMP the data out.
There's an additional important note I'd like to add though: mirroring snapshots/volumes then backing them up will back up "something," but without additional moving parts in place you wont' necessarilly have consistent data. For example, suppose Exchange resides on a VM living on the source volume. If this is all you're doing, you will get "something" at the destination, but it won't necessarilly be good. "CIFS" (SMB) should work out pretty well on its own for the most part, but VMware+NFS scenario is just going to be crash consistent unless you take additional steps to quiesce on the source side.
