NetApp Snapshot Technology

steve_ross — Thu, 05 Jun 2025 07:34:46 GMT

I understand that a snapshot is simply pointers to the data on disk that was snapped. Does this mean that if I interact with the snap (for instance if I back up to tape from the snap) that I am performing reads against the actual data (the snapshot pointers direct me to the data that has not changed)? If there is a good doc you know of that describes the IO implications of interacting with snapshots please let me know.

Thanks,

Steve

Re: NetApp Snapshot Technology

adamfox — Thu, 03 Apr 2008 23:39:57 GMT

The short answer is yes.

The longer answer is that for every 4KB WAFL block that has not been updated in the active filesystem since the snapshot was taken, you will be reading from the exact same block as one would in the active fielsystem. What I can also tell you is that the native ONTAP dump command (which would also be used for an NDMP-based backup) gets a lower priority than protocol ops, but they will most likely be contending for the same blocks, and more importantly, the same disks. If they were contending to read the same blocks at the same time, then read cache kicks in and tries to avoid making subsequent pulls from disk so you'd be less affected in that scenario. It's when they want vastly different blocks that you end up contending for disk I/Os would probably be a bigger potential performance issue. But there are stratgegies to help that as well, especially in sequential reads like backup (think aggressive read-aheads when one side does get the disks)

The best paper I know of is still the old tried and true TR3002: which is probably 15+ years old, but still has meaning. I would be wary of any limitation #s in there as things have obviously changed in the last decade or so, but it still explains the concepts pretty well (note it even still talks about an "NFS File Serving Appliance", which is clearly dated). I'm not sure if it addresses your exact concerns, I'll admit to having not read it in quite a while, but it's a good explanation of how it works.

The important thing that WAFL does that most other snapshot technologies to not do is it keeps existing blocks IN PLACE, rather than copying them out to a separate area. When a block is slated to be over-written, rather than copying it out, then over-writing the block, WAFL allocates a new block and adjusts the pointers. It's really efficient in that respect and you don't pay the signficant penalty the the other designs get when you deploy snapshots (I've seen some systems crumble at 5 or 10 snapshots, where WAFL keeps on rolling at 200+).

Hope this is at least a good start to answering your question.

Re: NetApp Snapshot Technology

steve_ross — Thu, 03 Apr 2008 23:45:09 GMT

WOW...what an outstanding answer. Adam, I am very appreciative of this detailed answer. It really addresses my question perfectly. Thanks for taking the time to answer so well.

Steve

topic Re: NetApp Snapshot Technology in ONTAP Discussions

NetApp Snapshot Technology

Re: NetApp Snapshot Technology

Re: NetApp Snapshot Technology