Network file systems

David Bandel david.bandel
Wed May 30 14:15:07 PDT 2007


On 5/30/07, Michael Hipp <Michael at hipp.com> wrote:
> Net Llama! wrote:
> > On Wed, 30 May 2007, Michael Hipp wrote:
> >> Is there a "network file system" that has all the features of a true Linux
> >> file system (e.g. ext3)? I'm thinking of features like inodes, hard links,
> >> soft links, unix security model, mtimes and atimes, journalling.
> >>
> >> I know samba pretty well and it doesn't do this AFAICT as it basically
> >> emulates a FAT32 fs over the wire.
> >>
> >> Does NFS do this? Is there some other "n. f. s." that does? (I read opinions
> >> of NFS not being all that stable.)

I've used NFS for years (have a network of systems running in a
school).  No problems on the stability side.  However, it is not that
secure.  I combine NFS with NIS so there are no username/UID
collisions.

> >>
> >> Can you do RAID over the network?

Umm.  Believe I read that that's possible, but I have no need.  All my
systems are RAID 1.

> >
> > I don't understand what you're trying to accomplish.  NFS (and every other
> > networked filesystem that I've ever used) has no understanding of the
> > underlying disk structure.  It sits on top of the real filesystem on disk,
> > which has inodes, etc.
> >
> > What is the problem that you're trying to solve?
>
> Sorry I wasn't clear.
>
> There are probably 2 things most important:
>
> - The ability to hard link files on a "network drive" ... mount the drive on a
> Linux box; copy a bunch of files there (cp -a foodir /mnt/nfs); create a hard
> linked copy of those files (cp -al /mnt/nfs/foodir /mnt/nfs/foodir1). Saves
> lots of storage space when you want an exact "time snapshot" of a bunch of files.

You cannot hard link files that are not on the same physical
partition, so that rules out NFS.

>
> - I'd like to be able to manage the security/access of those files across the
> network using standard nix tools chmod/chown. (I can probably work around this
> requirement.)
>
> - The other thing is that I use the Unison file synchronizer to sync some
> large directories. Unison first attempts to use inode numbers as a quick way
> to determine if a file has changed and needs to be synced. Failing that it
> falls back to the (much slower and more I/O intensive) rsync algorithm to do a
>   full file scan. I'd like to avoid that second step in the 98% of cases where
> it is unneeded.
>
> And I need to be doing this from more than one client system so the file
> system in question needs to be shared/exported.
>
> (The more mundane features like mtime, atime, journalling are probably taken
> care of by the underlying fs.)
>
> Does any of that make any sense?

Yes, but what you are trying to accomplish will not work (at least not
the hard link/unique inode across systems stuff).

NFS allows network access to a file system.  Each filesystem
(partition) has unique inodes, but they are not unique from one
partition to another.

Ciao,

David A. Bandel
-- 
Focus on the dream, not the competition.
            - Nemesis Air Racing Team motto



More information about the Linux-users mailing list