[212] in Info-AFS_Redistribution
Re: Problems with AFS 3.1
daemon@ATHENA.MIT.EDU (Torben Nielsen)
Mon Jul 15 17:44:11 1991
From: Torben Nielsen <torben@foralie.ics.hawaii.edu>
To: Info-AFS@transarc.com, Marybeth_Schultz@transarc.com, torben@hawaii.edu
Date: Mon, 15 Jul 91 11:01:31 HST
Marybeth,
>1- Transarc has had reports of problems with the r* commands. We are in
>the process of attempting to duplicate them. At this time, I cannot
>provide further information on the status of the r* commands but I
>will keep you posted on the progress.
Take two Sparcstations running SunOS4.1.1 and install all of the AFS
versions of the ``r" commands. Plus of course the AFS ``login". And you
should see the problem right away. We'll give you the ``inetd" debugging
output, but it won't do much good. It says nothing revealing at all. It's
more like ``someone wants to login" followed by a fork and a note that the
child has been reaped. And that's exactly what happens :-( I'm pretty
sure the problem lies in ``rlogind", but I've said that before.
>2- Yes, the current volume name limit is 22 characters (since space
>is reserved for the .backup and .readonly extensions). This limit
>will be increased in DCE AFS. It is not a trivial matter to change
>this limit in AFS 3 because it is buried in volume headers, the vldb,
>etc. We are currently evaluating what would be necessary to increase
>this limit in AFS 3.
That doesn't sound good. Unfortunately, I believe it is a real impediment to
setting up large trees. Please let me know when you reach a conclusion. This
one is serious enough that we'll have to go try it ourselves if nothing
else and I'm really loath to do that. A value like 256 would be pretty
good for now. Hey, even 64 would be decent, but I don't understand why it
has to be so small. What I was really worried about was that maybe someone
had a cute algorithm in there using bit manipulations and depending on
32-bit words :-)
>3- On AFS clients running SUN OS, the output of "ps aux" displays the
>afsd processes without names. The "ps" command allows you to
>either get names or have messed up load averages. The latter has
>real problems since alot of programs defer doing any work until the
>load average falls below N. As a result, Transarc has chosen not
>to display the names.
Hmm.... How about Sun's ``biod"'s? Those processes have names and they're not
worried about screwing up the load average.... Which programs depend on the
load average by the way? The only example I can think of off the top of my
head is ``sendmail"..... Seriously, I don't see why you can't just do like
Sun does with ``biod"..... Those do not raise the load average.
>4- The pathname "/afs/transarc.com/afs.rel" is no longer a valid
>pathname. Please use "/afs/transarc.com/systemtype/usr/afs.rel".
>The AFS Fact Sheet will be corrected.
Thanks; that was a minor nit. But it's nice to get it corrected.
By the way, in case it sounds like I'm being a bit ``negative" about AFS,
that really isn't the case. I think it has tremendous potential and I like
it overall. Unfortunately, I keep running into problems that prevent me
from doing reasonable things :-(
Cheers, Torben
