lkml.org 
[lkml]   [1998]   [Mar]   [14]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
SubjectRe: File system snapshopts: how valuable?
Date
Colin Plumb wrote:

> I'm designing another file system, and I've stolen a lot of ideas from
> Netowrk Appliances' WAFL (write anywhere file layout), which you can
> see all about at http://www.netapp.com/technology/level3/3002.html.
>
> One feature they have which is rather cool, is snapshots: they
> can take a static copy of the file system at any time and keep it
> on line as a backup. Further writes are done via COW techniques;
> new data is written to new blocks and the old data is preserved.
>
> The only problem is that this complicates things like memory-mapped
> files (which NetApp's NFS server doesn't have to worry about).
> I'm wondering whether it's a valuable feature worth trying to
> implement or not.
>
> I suppose big-linux would be a better place to ask, but does anyone here
> have any opinions? If it were free, obviously I'd put it in, but
> there's significant hassle involved in some of the bookkeeping.

Very valuable, yes, but to a rather small audience. I presume (not knowing a
lot aboutWAFL and such) that there are rollback/commit-style calls you can
make to "fix" newly written data in ways similar to what relational databases
usually do. I suppose it could be useful for partitions containing users home
directories and stuff, given that the appropriate system calls could be
wrapped in nice tools. Kind of how those trash cans should really work in
other OSes. Can't think of any other large uses, but yes, it might be useful.
Possibly for programs handling enormous amounts of data stored on disk,
wanting undo capabilities; only... what programs?

Problem is, that partition will be problematic for other purposes, most
notably for LRU-like swapping, where a lot of writes would generate large
amounts of COW. So you would have to have a mechanism for "flushing" blocks
at the bottom end of a chain of COWed blocks.

Quest


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu

\
 
 \ /
  Last update: 2005-03-22 13:41    [from the cache]
©2003-2011 Jasper Spaans