[ih] Distributed file systems [was: As Flag Day approaches at CMU]
Barbara Denny
b_a_denny at yahoo.com
Wed Sep 10 20:41:01 PDT 2025
I was curious about the potential influence of the Unix United paper on work at CMU so I decided to just ask Satya. In my message to him, I included the part starting with the sentence "AFS in particular *must* .." and the 2 examples that follow from Brian's email.
I did tell him I wanted to post his answer to this list and he hasn't said no so ...
Excuse my trimming of the thread. I seem to have problems posting to the list, especially when the body of the message is long.
barbara
----- Forwarded Message -----From: Mahadev Satyanarayanan <satya at cs.cmu.edu>To: Barbara Denny <b_a_denny at yahoo.com>Sent: Monday, September 8, 2025 at 01:45:16 PM PDTSubject: Re: AFS question
Hi Barbara,
The Unix United paper (aka "Newcastle Connection") was published in
1982. We were indeed aware of that work by mid-1983, when serious
work on what eventually led to AFS began. The name "Andrew" for the
whole project did not emerge until late 1985. In fact, the first
published paper on AFS did not even use the name "AFS". It referred
to the system as "The ITC Distributed File System" and the server
and client components as "Vice" and "Virtue" respectively.
Here is that very first AFS paper, from 1985:
https://dl.acm.org/doi/10.1145/323647.323633
Alas, the influence of Unix United on AFS was quite the opposite of
what the person who spoke with you thinks. We actively worked hard to
AVOID an aspect of Unix United that we thought was totally wrong.
>From the beginning we believed that AFS needed to have Location
Transparency. i.e. you could not tell where a file was located by
just looking at its pathname. You had to ask the system to tell you,
and that location could change over time. The 1985 paper explicitly
contrasts AFS with the design of Unix United. If you look at
Section 6.1 in the above paper, it says this:
>
> Location transparency is a key issue in this context. In Locus,
> Vice-Virtue, Apollo and Roe it is not possible to deduce the
> location of a file by examing its name. In contrast, the Cedar File
> System and the Newcastle Connection embed storage site information
> in pathnames.
>
Location transparency was identified as a non-negotiable requirement
of AFS since the very earliest conception of its design. See, for
example, this September 1983 design document:
http://reports-archive.adm.cs.cmu.edu/anon/itc/CMU-ITC-008.pdf
BTW, there is a whole treasure trove of very early (1983-1985) design
documents from the ITC (Andrew project) at
http://reports-archive.adm.cs.cmu.edu/itc85.html
Many things evolved over time, of course, but these early documents
capture the state of thinking at the time they were written.
So the answer to your colleague is "Yes, Unix United was a big
influence on AFS, but in a totally negative way". You may wish
to soften the blow in how you present it to him/her :-)
Cheers
--- Satya
On Saturday, September 6, 2025 at 06:55:51 PM PDT,
Brian E Carpenter via Internet-history <internet-history at elists.isoc.org> wrote:
"MIT wasn't the only place where the IMP became the de facto local area net."
Which reminded me of Scrapbook at NPL. It was an early hyperlinked system but was also a (small scale) distributed file system by the mid 1970s. It was not widely known and is badly documented.
I happened to meet and interview one of the Scrapbook team last year:
https://www.cs.auckland.ac.nz/research/groups/CDMTCS/researchreports/download.php?selected-id=884
The Andrew File System and its descendants, like NFS, only came along in the 1980s. AFS in particular *must* have been influenced by the Unix United paper, which had examples like:
cd /../unix2/user/brian
quicksort a > /../unix1/user/brian/b
(where unix1 and unix2 were host names, and brian wasn't me, it was Brian Randell.)
Regards/Ngā mihi
Brian Carpenter
More information about the Internet-history
mailing list