[OpenAFS] Re: Moving large data from volume to volume
Mag. Mike B. Kerber
Thu, 13 Oct 2011 16:08:10 +0200
On 10/13/2011 03:53 PM, firstname.lastname@example.org wrote:
> To: email@example.com From: Andrew Deason
> <firstname.lastname@example.org> Date: Wed, 12 Oct 2011 12:49:14 -0500
> Organization: Sine Nomine Associates Subject: [OpenAFS] Re: Moving
> large data from volume to volume On Wed, 12 Oct 2011 11:51:42 +0200
> "Mag. Mike B. Kerber" <email@example.com> wrote:
>> > Hello!
>> > While reorganising our afs tree layout i was wondering if there is a
>> > on-server possibility to move large data set from a volume to another
>> > one (both RW on the same server) without sending the data itself over
>> > the wire.
> No (at least, not without fiddling with the internal on-disk
> structures). However, there is some code for the functionality of
> "splitting" a volume into two volumes at a directory boundary, though it
> is not yet fully functional in the stock OpenAFS source. It doesn't seem
> like it would be too difficult to add the opposite functionality:
> merging two volumes together. If those two operations existed, you could
> have a way to move data like that, although it is perhaps a bit clumsy.
> Or we could just make an operation to move/copy a file from one volume
> to another if they're on the same partition. But nothing exists like
> that now. (Naively, it seems like RXAFS_Rename could be made to do this,
> You can of course issue the move from a client running on the
> fileserver, which may reduce some of the network overhead, but that's
> still much slower than a rename().
>> > Judging from performance this seems to be what happens...
> -- Andrew Deason firstname.lastname@example.org
thanks for the answers and suggestions. For me playing with the on-disc
structures is a no go ;)
It is cool to view the raw data but thats about it.
To move along directory borders is exactly the case here, as the tree
gets reorganized and part of the old volume gets moved as a whole into
They are on the same partition even, but i think even a move from
partition to partition would be a huge speedup.
If this could be managed invisible to the user that would be cool, even
doing it with an 'fs' command or similar would be nice.
For now i will wait until all the data went around.
Anyway thanks for all the work!