[OpenAFS] OpenAFS on MacOS X

[Derek Atkins]

"Patrick J. LoPresti" <patl@curl.com> writes:

> Ah, right (duh).  What I meant to ask was whether the communication
> was with the file server or with some other replicated server (e.g.,
> the VLDB).  Either way would give you problems in the top-level /afs,
> but at least within a cell, the VLDB should always be available even
> when particular file servers are not.

I just ran a network trace to be sure that my answer was accurate.
The stat() will indeed contact the VLDB (if necessary) and then it
will contact the FileServer directly.

> I am concerned about the case where one of our file servers is
> unavailable and someone tries to browse, say, /afs/curl.com/users.
> Just how bad is the timeout, anyway?

Timeout for each down server would be relatively short (a minute?).
My suggestion would be to break up "users" into smaller subdirectories
like users/[a-z]/...  Besides, if you have a downed fileserver you
have bigger things to worry about than people trying to browse the
user list :)

> Having an option to make stat() on mountpoints not chase them would be
> handy; you could solve this problem with replicated RO volumes.

The problem is that then you could not trust the information you get
back from stat().  The owner, unix permissions, etc. are all stored in
the volume itself and not in the mountpoint.  You can't easily change
_that_, because volumes can be mounted in multiple places.  It would
violate the security semantics if a volume took the permissions of its
parent.

-derek
-- 
       Derek Atkins, SB '93 MIT EE, SM '95 MIT Media Laboratory
       Member, MIT Student Information Processing Board  (SIPB)
       URL: http://web.mit.edu/warlord/    PP-ASEL-IA     N1NWH
       warlord@MIT.EDU                        PGP key available