>From: "James A Goodwin" <jagoodwi@xxxxxxxxxx>
>You were right about the -D option, the xfsrestore works fine now.
>As for using the -a, it is not important. I think that DMF must work a bit
>differently than HPSS. We actually punch holes in the file after migrating
>the data. I guess that DMF does not, which requires use of the -a option
>to xfsdump in order to keep the migrated data from being dumped. For us
>the data isn't there at all, so xfsdump doesn't have a choice to make.
DMF might punch, or it might not. It's possible for the file to be online,
and to also have a valid offline copy at the same time. We call this a
dual-state file. When DMF is otherwise idle, it sometimes starts picking
candidates of least-recently-accessed files and copying them out to tape,
making them dual-state. Then we don't punch the files until we absolutely
have to punch them.
These dual-state files allow DMF to stay ahead of the game, so to speak. When
someone starts to quickly suck up disk space, and we need to quickly make room
to keep the filesystem below the high-water mark, we just have to punch the
dual-state files. We don't slow down the users by making them wait while we
copy some files out to tape.
On the other hand, if they decide that now they're going to access one of
these dual-state files, well...it's already online, so they don't have to wait
for us to copy it in from tape.
So, the -a option tells xfsdump that it should not try to read the file when
it has a valid off-line copy of a file that is also online (why make yet
another backup of the file when it's already on tape?)