Subject: Re: New submission: RFS
To: Greywolf <>
From: Matthew Jacob <>
List: tech-kern
Date: 10/20/1999 21:03:21
On Wed, 20 Oct 1999, Greywolf wrote:

> On Tue, 19 Oct 1999, Greg A. Woods wrote:
> # However even AT&T's original RFS would still need some major overhaul in
> # order to work well on a modern BSD, so I'm only lightly suggesting that
> # the name "RFS" be reserved for it or something like it (i.e. a stateful
> # remote file sharing protocol that preserved "Unix" filesystem semantics).
> # 
> # For example the SunOS-4.1.4 implementation demonstrated this in that
> # some operations were not supported fully, and some data types were not
> # fully transportable (such as an inode number, restricted to "ushort" in
> # SysVr3). 
> Keep in mind that RFS was never designed to be used in an OS-wise hetero-
> geneous environment.  AT&T intended it to be run in an AT&T SVRx environment
> and damn the torpedoes.

Ah, not entirely so. It in fact was offered to Sun as an alternative to
NFS. There was a serious amount of thought and consideration given to it,
as well as negative words. Howard Chartock actually made it work in 4.X,
at about the same time Guy Harris was running screaming down the hall
getting streams in 4.X. It eventually became owned by Chia P'Tien (my
former officemate) where it languished and went away- probably before even
Solaris 2.1 shipped (I'd left by then...).

While it didn't have the XDR/RPC base of NFS which gives NFS more OS
heterogeneity, it did manage to do remote device access stuff for those of
us who felt that RMT protocols and their variants really aren't the right

> # Some of these issues were dealt with in SysVr4 though, IIRC. [I used RFC
> # with SunOS-4.1.4 to an ISC-2.2 (SysVr3.2/i386) box once and was
> # successful in accessing tty devices for remote modem sharing....]
> I think this was fortuitous happenstance.  SVR4 was the first System V
> to begin to catch up with the rest of the UNIX world (remember:  most
> sites that used networking were, for a LONG time, based on Berkeley-style
> systems -- including the Berkeley FFS.  SVR4 finally caught on to this
> and used it.)  There were hybrid extensions (UniSoft UniPlus+) which
> were SVR[23] with NFS and NIS glued on, but...

Ah yes, UniScum...... I will have to say, though, that in the early 80's
there were many of us who found Berkely Unix too much of a grad school
project- and for all of the faults in, say, System III or SVr2 (for which
I did Dual Systems' 68020/68881 compiler), there was at least *some*
attempt to address commercial needs which, frankly, it took Sun until
about 1989 to even begin to get the BSD source base pointed toward. Then
there was the buyback of Sun code into BSD4.4 (including specfs bugs that
I had in fact *FIXED* in SunOS 4.0.3 but still showed up in BSD4.4 and
then into OSF/1), and the rest, as they say, is Hystory....

Perhaps some of my disdain of Berkeley comes from them ignoring my fixes
to 2.8/2.9 for 22-bit Qbus LSI-11/73 fixes... Oh well, such is life...