afs: considered harmful (was: Re: URI WG decides a couple of things)

"Peter Lister, Cranfield Computer Centre" <P.LISTER@mail.cranfield.ac.uk>
Message-id: <9307131132.AA14336@xdm039>
To: uri@com.bunyip, www-talk@ch.cern.nxoc01
Cc: ccprl@mail.cranfield.ac.uk
Subject: afs: considered harmful (was: Re: URI WG decides a couple of things)
In-Reply-To: Your message of "Tue, 13 Jul 93 09:20:42 BST." <9307130820.AA04227@xdm001>
Date: Tue, 13 Jul 93 12:32:16 BST
From: "Peter Lister, Cranfield Computer Centre" <P.LISTER@mail.cranfield.ac.uk>
Status: RO

> 1. The afs: URL type was removed from the spec due to lack of interest
>    and because it had not been really discassued or thought out.

Hooray. Speaking as an AFS administrator, I always thought it rather
odd. I thought I mailed some comments on it to Tim BL, but maybe these
got lost in the post (UK mail is weird, all the more so because UK
mailers try to treat CERN as part of the UK). So, just to hammer the
nails in the coffin, I'll repeat my objections.

1) AFS sites access another AFS site's files simply by treating them as
local files, so the existing file: URL type deals with this. AFS sites
following the normal naming conventions simply share onw massive
"local" filesystem. The problem of exactly what file: refers to in
respect of the local system has been discussed elsewhere, but AFS
avoids this, as no file server name appears in the URL, except
conventionally as part of the filename, e.g.
file:/afs/pegasus.cranfield.ac.uk/home/cc/ccprl, NOT
file://pegasus.cranfield.ac.uk/home/cc/ccprl.

2) AFS sites are under no obligation to mount volumes consistently in
one place so that a particular pathname will always work. Though in
practise, most sites do, there are potential problems if an afs: URL
assumes that a set of files which haven't moved in AFS terms are
mounted in several different places in the AFS tree at once. Naive
applications will probably see several copies which mysteriously appear
and disappear. "Mounting" in AFS terms, can be done by anyone,
anywhere. One could make the contents of the afs: URL into an AFS
cell/volume pair and a relative filename, rather than a pathname, but I
can't honestly see any point.

3) AFS sites can only see other AFS sites via AFS. Fairly obvious, but
the point is that an AFS site wishing to publish information widely
will do so in other ways (anon ftp, http, etc). AFS is also restricted
to U**x clients, which rules out all PC people, even if they're at AFS sites.

4) AFS is not well suited for read-only access of freely published
material. Compared to http, there is a large overhead in terms of
authentication, authorization, cache maintenance and so on. These are
great for actively sharing files amongst a small group of people, but
vast overkill for most WWW stuff over the wide area. Read-only volumes
help a bit, but not much.

Just out of interest, who is using AFS in the www community?

Peter Lister                                    p.lister@cranfield.ac.uk
Computer Centre,
Cranfield Institute of Technology,        Voice: +44 234 754200 ext 2828
Cranfield, Bedfordshire MK43 0AL UK         Fax: +44 234 750875