On 09/04/99 13:23:47 you wrote:
On Fri, 3 Sep 1999 sirbruce@ix.netcom.com wrote:
So to change the ID requires essentially a recursive chown-type operation that is always going to be proportional to the amount of data "schlepped".
If it really was as fast as a recursive chwon, that would be great, but the Netapp still issues a copy/unlink instead of a rename when moving files between quota trees. This isn't the fault of the client either, because the same thing happens if I mount the filer's root filesystem on the client (so it thinks it is one filesystem), and do a move from there.
Hmm, this doesn't make any sense to me. Shouldn't do that if the client has the root mounted, and in any case I don't know how one would fine out what ops the Netapp is "issueing" to itself. Perhaps it's reported in the NFS ops mix? But if that's the case, then you have an issue as to whether or not those commands should report only what the client is sending (my view) or what operations the Netapp is translating that to internally (your contention that it does now).
How about an official "wafl_stat" command?
Bruce
On Sat, 4 Sep 1999 sirbruce@ix.netcom.com wrote:
Hmm, this doesn't make any sense to me. Shouldn't do that if the client has the root mounted, and in any case I don't know how one would fine out what ops the Netapp is "issueing" to itself.
When the client tries a rename() between two qtrees, it fails with EXDEV, so the mv command falls back to a copy, even though it should appear as a rename between two directories on the same filesystem (in the case of a root-mounted filer). Yes, it would be great if the filer could twiddle with its file table and be able to do true renames between qtrees.
But if that's the case, then you have an issue as to whether or not those commands should report only what the client is sending (my view) or what operations the Netapp is translating that to internally (your contention that it does now).
I pinned the blame on the wrong end of the connection. ;-) The NFS client is fully aware that it is issuing reads/writes to copy the file, and the NFS stats reflect this.
-
Brian Tao -
sirbruce＠ix.netcom.com