tlug: Re: AFS -vs- NFS (was: Office suite for use under Linux)

[Rex Walters <rex@example.com>]

On Mon, Apr 20, 1998 at 07:38:56PM +0900, Stephen J. Turnbull wrote:
>
> .... Andrew also comes up in
> a different context as the Andrew File System, a distributed file
> system (in the sense that NFS is) which scales better than NFS to
> large complicated networks.  I'm not sure why it has never caught on.

Well, I can certainly take a stab at *that*!  :-)

IMHO its because virtually every Unix vendor bundled support for NFS,
whereas AFS required you to add an additional (commercial) product
on top of the base OS.  Very similar to M$ having vendors bundle
Win3.1/Win95/Tomorrow-The-World with every PC they sell.  

[On a side note: anybody remember RFS?  Or even more unrelated NeWS? :-) ]

NFS was never really open source until recent free Unices (although the
protocols have been documented as RFCs for a while).  AFS, AFAIK, has
never been open source since the earlyearly CMU days (i.e. since Transarc
started selling it).  If anyone knows about free/open implementations
of AFS or DFS I'd be very interested in hearing about them.

AFS has many nice technical features over NFS (read-only replication,
interesting volume management, backup/snapshot directories, among
others) but can be painful to implement in an existing large network
(no bidirectional compatibility with NFS among other problems).  IMHO it's
also slow (excpet for read-mostly data) and too complicated/fussy to tune
performance.  Many folks also dislike a single vendor solution (truly
single vendor since IBM bought Transarc).  Finally, transitioning from
AFS to DFS is every bit as difficult as transitioning from NFS to DFS
(AFS has no future other than DFS if that can be called a future).

The only places I've seen AFS these days in the US are Universities,
SAS Institute, Intel, and IBM itself (all of which are unusual in that
they have large supplies of extremely Unix-savvy brainpower -- many other
places really *are* run by pointy-haired types).  In Japan, I think
Honda is a big user (as well as several Universities).

I like a *lot* of the features of AFS and DFS, but I'm now of the opinion
that "transition is everything".  If it isn't easy to get there from here,
people won't go there.  The path of least resistance will always be followed.

I suspect that most of the really good ideas from AFS, DFS, etc. will
eventually make their way into NFSv4, NFSv5, and onward.  There is a lot
to be said about a nice, simple, stateless 16 operation protocol like
NFSv2 (most of the problems over the years have been with the associated
lock/mount protocols and not with NFS itself).  Kind of like POP -vs-
IMAP:  IMAP is simply better but most of the real world still uses
POP (for a variety of reasons).  (No, I'm *NOT* defending some of the
silliness in POP!).

Filesystems (local or network) are fascinating beasts.  No one solution
is likely to satisfy all comers.

What all this has to do with Linux and Open Source Software is beyond me,
but it hits a nerve near-and-dear to my heart (I keep my wallet in my
breast pocket!).  I like to keep an eye on the major trends in protocol
wars, looking for the next HTTP (or the next OSI/GOSIP for that matter!).

Cheers,
-- 
Rex
---------------------------------------------------------------
Next Nomikai: 15 May Fri, 19:30 Tengu TokyoEkiMae 03-3275-3691
Next TLUG Meeting: 13 June Sat, Tokyo Station Yaesu gate 12:30
Featuring Stone and Turnbull on .rpm and .deb packages
---------------------------------------------------------------
a word from the sponsor:
TWICS - Japan's First Public-Access Internet System
www.twics.com  info@example.com  Tel:03-3351-5977  Fax:03-3353-6096