Hi all,
A question regarding verify passes on NDMP back-ups.
We're currently performing weekly back-ups of our FAS2050 to tape using Backup Exec 12.5 + NDMP option, with the tape drive attached to the Backup Exec media server (rather than the filer itself) as we're also backing up non-filer data.
The back-up itself executes, and we then run a verify pass to verify the data on the tapes.
We are experiencing the sequential-reads-are-very-slow issue in our release of Data ONTAP (7.2.4L1), which is resulting in the backup pass only running at 15MB/s or so. Annoying, but we know a fix is in the works. I noticed, however, that the verify pass runs at closer to 50MB/s, and was investigating why this is as I would have expected it to be similarly slow.
Looking at sysstat on the filer while the verify pass is running, a large amount of data is being received at the filer, whereas I would have expected it to be sending a large amount of data. I can only assume this is because the verify operation sends the data from tape -> server -> filer and the verify happens at the filer, rather than than data going from tape -> server and filer -> server and the verify happening at the server as I was expecting? This would help to explain the better performance during the verify pass as well.
What I can't expain, however, is this: during the verify pass the filer is receiving data at 40-60MB/s on it's network interfaces. However, the Disk Read stats in sysstat are much lower: 1MB/s - 5MB/s (which is our "normal" disk read load). This makes me wonder if the data from the back-up isn't actually being verified?
Can anyone explain this? Is the filer only pretending to verify the data (so it's just wasted effort to run a verify pass)? Or does the filer verify the back-up data based on checksums/metadata rather than byte-by-byte, which therefore would require very little in the way of actual disk reads?
Thanks,
Matt
