Subject: Re: AFS/arla
To: Gary D. Duzan <>
From: None <>
List: current-users
Date: 09/14/1998 13:35:59
"Gary D. Duzan" <> writes:

> =>I suspect that the CMU folks will now chime in with more information.
>    From the looks of the source-changes traffic, it is still very
> new and very much in flux. I'd seriously doubt if it is usefully
> functional at the moment in current. Soon, perhaps. In any case,
> having it in the tree is a positive sign. I've thought Coda was a
> cool idea since SIGOPS 13.

Let me give a CMU answer.  First, yes the tree is changing a lot in
the past few days (and sadly, the app part of the tree [which is only
at the coda web site] was just made slightly inconsistent with regard
to the kernel -- this should get straightened out tonight and

BUT THE REAL answer is that Coda is only in flux lexically.  Coda has
been around in 1.3 and (freebsd 2.2.6 and linux) for a pretty long
time now.  It was ported up to -current a months or so ago and
installed into the tree.  In that process, people suggested that
calling it cfs was possibly ambiguous and hence the most recent
thrashing in the sys area.  I also plan to install the coda package
into the /usr/pkgsrc area, imminently.  I was just talking to
Allistair as to whether he would object to my making a new package
category, filesystems, for coda and others.  (comments?)

So, I would never say that there are no bugs in Coda, but Coda is
definitely usable in a simple client/server operation mode.  It also
allows for disconnected operation (with laptops) which is quite handy.
But the biggest thing that Coda has going for it is that it is free
and all the source is available for scrutiny and debugging/enhancment/...
Past experience with free public systems (gnu, X11R6, linux, netbsd,
freebsd, apache ...) seem to have proven the case that they might
start out shakey, but they gain critical mass and nuke the

(Someone raised the comments that Coda only works on pc's.  Clearly,
we don't have the hardware and expertise to make it run on all 
platforms.  But the kernel proper should be platform independent.
The changes to the app part should be minor -- coda has a light weight
process scheme that requires some asm code.  Phil can show us how
hard it is/was to get Coda on an arm32.  Once Coda works on 2 
platforms the jump to n shouldn't be much of a technical challenge.)