hi,
i am testing ndmp backup solutions on my f760[part of a cluster] running 5.36r2, against a spectralogic treefrog. questions:
- once backup started sysstat 1 showed cpu util at about 40% during mapping and
initially 100% during dump for about 2-3 minutes and then much lower numbers since then. is the 100% util normal? nothing else is running, system is not prod yet.
I can answer this question.
At the beginning of any BSD format dump, there are a couple of bitmaps. One bitmap has a bit for every active inode in the subtree, at the time of the snapshot you're dumping. The other bitmap has a bit set for every inode on the dump tape.
On a level 0 dump, these bitmaps are the same. On an incremental dump, they are probably different (if a file hasn't changed, it is still active, but it won't be on the level 1 dump tape).
Anyway, the mapping phase builds up these maps. Then, at the beginning of the dumping phase, we slam the maps onto the tape.
Since your system in inactive, the entirety of the maps are probably in buffer cache. So we're just reading from memory and writing to tape.
Once that is done, we start reading the directory data from the file system. That is not all in buffer cache, so CPU usage drops some, and disk activity should jump.
On an active system, the bitmaps will probably not all be in cache. Furthermore, dump effectively "nices" its CPU consumption. It won't hog the CPU if NFS/CIFS etc. operations are in the queue. You may still see CPU hit 100% for 2 minutes at that stage in the future, but that's just dump using whatever CPU that the rest of the system has left over (why waste it? :)
Good luck, Stephen Manley DAM and NDMP Steve Nash Hairstyle Fan
- can ait2 tapes be "bad" fresh out of the wrapper? the software would not
handle 1 of the tapes, backup vendor said tape was probably bad. what is the ndmp response to a "bad" tape?
- is the chip in the ait tapes used by ndmp backup?
thanks