[OpenAFS] OpenAFS in a production environment

Jeffrey Hutzelman jhutz@cmu.edu
Wed, 07 Sep 2005 18:29:14 -0400

On Wednesday, September 07, 2005 09:49:54 -0400 Derek Atkins 
<warlord@MIT.EDU> wrote:

> "Todd M. Lewis" <utoddl@email.unc.edu> writes:
>> Lester Barrows wrote:
>>> In an out-of-band discussion, Jeffrey Altman has managed to convince
>>> me (not an easy task mind you!) that Transarc AFS servers are more
>>> likely the cause of our NAT troubles with AFS clients. For this
>>> reason I'm amending my initial statement to say that if you access
>>> Transarc AFS servers, you should not put AFS clients behind a NAT.
>>> This is apparently due to IBM/Transarc AFS servers' UUID tracking
>>> not behaving correctly when multiple clients come from the same
>>> IP. [...]
>> As long as you brought it up, IBM/Transarc AFS servers don't behave so
>> hot when the (apparently) same client suddenly shows up coming from lots
>> of different IP addresses either. I don't know whether OpenAFS servers
>> deal with it any better.
> They do.  Jeff Hutzelman and I worked on a fix for this at the IETF
> in Minneapolis about three years ago.  It was fixed in a middling 1.2
> release.

What we fixed was crappy behavior when a client falls off the network and 
then later comes back with a different IP address, while the fileserver has 
tried to break a callback in the meantime.  The IETF network had some 
problems that week, one effect of which was that clients tended to do this 
a lot.

I think the problem Todd is referring to is what happens when multiple 
unrelated clients all claim the same UUID.  I imagine we probably don't 
deal with that too well.

-- Jeff