[ale] School Project to Create Distributed Filesystem

[George L. Allen]

On Tue, Feb 24, 2009 at 10:57:27AM -0500, Greg Freemyer wrote:
> 1) GFS from Redhat.  A cluster FS.
> 2) OCF2 from Oracle. A cluster FS
> 3) ZFS a filesystem from Sun.  Offers mirroring I think.  distributed?
> 4) BTRFS another new generation FS with checksumming.  Not sure if it
> has any relevant aspects or not.
> 5) DRBD from linbit(?) - a distributed raid-1 kernel patch. ie. half
> the mirror on one computer, half on the other.
> 5.1) Linux-HA - cluster enabling code typically used with drbd.  Has a
> shiny new drbd/linux-ha gui available (as of a couple weeks ago).
> 6) CFS from OpenSSI - a cluster FS
> 7) UnionFS - A layered filesystem the sits above other FS's. I would
> look hard at this if I were you.
> 8) Lustre - A true distributed FS that may satisfy all of your goals
> already.  So maybe you want to avoid Lustre?  Or if just a goal or two
> are missing, maybe you could enhance it?
> 

Has anyone actually used GFS, Lustre, etc... for a setup such as this??

For example - given two datacenters at different ends of the state, and a
few dozen remote sites on T1's, I'd like everyone (windows users) to be
able to mount a user-dir, from their closest server, but have the data
become redundant across the other servers also. Preferably through
live-mirroring (forgot the buzz-word term for that) rather than
replication.

My first inclination would be to use AFS (Andrew File System) ... but I
don't believe it's replication meets the requirements of having mirrors
with real-time locking rather than replication.

My second inclination would be the Solaris with per-user zfs+SMB-shares,
but from what I've read so far, the real-time replication/mirroring (via
iSCSI?) is not yet free - and is still part of Veritas?

But the big question is which of the other half dozen filesystems already
does this well.