We're using NDMP to FC tape for weekly fulls, and NDMP to a disk staging STU on the master server for daily differentials.|
It's very reliable. The only issue I have is the performance. For the FULLs, we write to 2 tape drives in parallel. We get about 50-60 MB/sec to each drive. I know the drives are capable of a lot more - our Linux media servers can pump out 160 MB/sec to a single drive. But our Netapp box is relatively low end and quite old now (V3020 gateway - which has been superseded)
It takes just over 24 hours to do the weekly FULL backup (about 10TB). We run one backup stream per Netapp volume, of which there about 22, ranging in size up to 1.3TB. We don't multiplex to the tapes - writing one stream per tape drive.
I've messed around with multiplexing and the number of drives being used in parallel, but I've never been able to get it above that 50-60 MB/sec per drive.
The drives are IBM 3592-E05 (TS1120 is their other name, I think).
Doing the daily DIFFs to disk works well. Incremental backups can take a long time to decide what actually needs to be backed up before any I/O actually happens. Directing these to disk means NDMP is not holding a tape drive for a long time without actually doing any I/O.
NDMP to disk became possible with either 6.5.1 or 6.5.2
A nice thing is that ONTAP (the Netapp operating system) seems to be pretty good at prioritising "user I/O" over backup I/O. If our full backups run into online time, when we have thousands of users connected connected via CIFS for their desktop file serving needs, the backup speed will drop right down, as ONTAP prioritises the user's I/O. Annoying for the backups, but good overall for end user performance.
Veritas-bu maillist - Veritas-bu AT mailman.eng.auburn DOT edu