[OpenAFS] 2.6 linux kernel support; afs from ramdisk

[Dr A V Le Blanc]

On Saturday August 7, Christoph Hellwig, a Linux kernel developer,
gave a presentation at Leeds on Linux filesystems.  He said that
AFS was a buggy filesystem with a very messy code base, and that
people should stay away from openafs, which was particularly
unreliable.  I don't want to blame Christoph, who seems to have
picked up some anti-AFS feeling from some other kernel programmers,
but shouldn't we be doing something or other to counter this bad
press?  I stood up to say that we had been using AFS for about 15
years, and openafs since it came out, and that we were completely
satisfied with it.

Incidentally, I presented a paper at the same conference about a
distributed desktop system we are using here; the Linux base
system contains mainly a kernel and ramdisk, which can be booted
over the network.  The same kernel and ramdisk are used for
installation and for ordinary running of the system.  Installation
takes about 2.5 minutes.  The ramdisk mounts a root filesystem, does
a pivot_root, and starts AFS.  Then package is run, followed by
exec /sbin/init 2.  As a result we have a complete linux system
with 2.3 gb of files in AFS and only about 28mb on the disk,
not counting the 500mb in /var/cache/openafs.  The advantage is
that we have scalable cluster Linux that runs fairly efficiently
and can be administered approximately as a single machine.  We
have this currently installed on about 200 machines, but this may
grow to about 3000.

     -- Owen
     LeBlanc@mcc.ac.uk