Obviously from all the NDMP messages I'm sending this week, we just started using NDMP. :-)
I just saw the results of our first incremental NDMP backups and I was flabbergasted at how fast they ran. I even double-checked the logs to make sure it actually was backing stuff up.
When doing an incremental backup over CIFS, it would take about three hours to go through about 200 GB of stuff (1.8 million inodes). Doing it with NDMP, it takes 20-30 minutes. That's quite an improvement.
It's obvious that when doing an incremental NDMP backup, there is something going on that allows the filer to determine extremely quickly which files have changed since the full backup.
Can someone tell us what's going on? Does the filer keep a log of stuff that's changed? Is this just something WAFL is really good at?
Basically, the performance comes because our backup software is tightly integrated with the filesystem. And because WAFL is a really, really high performance file system.
We are able to do things with prefetching and file system examination that an application using NFS/CIFS just doesn't have access to or control over.
Just another benefit of the "network appliance" running WAFL.
Stephen Manley NDMP and Dump-man Emeritus