On 12/09/11 13:20, Jason White wrote:

Peter Ross<Peter.Ross@bogen.in-berlin.de> wrote:

...

Every snapshot is like a filesystem, with its own metadata and data.

They only "interfere" because they may point to the same (unchanged) data. But there is no backtracking or reconstruction from one snapshot to the other.

My understanding is that this is also true of Btrfs snapshots. It should be noted that most of the metadata are not copied when a snapshot is created, judigng by the speed of this operation.

Regarding ZFS, it is my impression that all of the people who understand the internals of the file system and its implementation are Sun/Solaris developers, some of whom may have left after the acquisition by Oracle. I don't know what situation this creates as to the availability of expertise (at the level of understanding the code well enough to debug it) for projects porting it to BSD and Linux.

Btrfs does appear to be gaining contributions from a variety of sources, prominently Red Hat, so the expertise isn't concentrated in the hands of any single corporation.

Last time I checked, Btrfs had sucky performance, especially for PostgreSQL loads. (Even with the "nodatacow" option enabled) How does the DKIM-based ZFS look? Otherwise I'm curious enough that I'll run up a test myself. Cheers, Toby