Why does automounter fail once then never work again?

[Tim Fredrick]

(Regarding AIX3.1.5)

We have had this situation since we first tried to use automounter.  We
are starting it up like this:

       /usr/etc/automount /r /etc/auto.direct

as the last thing done in /etc/rc.tcpip.  A couple of our our auto.direct
entries look like this:

	u1	-rw,intr,soft	pyr:/u1
	s1	-rw,intr,soft	wk1:/s1

These are a Pyramid 90x and a SPARCstation-1 (SUNOS4.1.1) respectively.
Everything seems to work just fine until the first mounting problem -- a
machine goes down, or something times out because of a busy network, etc.

Then our disasters begin.

Doing a "ps -efgaux" doesn't show automounter running in memory.  Yet a "df"
command will hang almost indefinitely sapping the IBM's CPU.  Any attempt to
cd to /r/s1 or /r/u1 does the same thing -- that process hangs and saps CPU
time.

Attempting to stop and start the nfsd and biods doesn't work.  Nor does
"stop nfs" and "start nfs" in SMIT work.  The only thing that we've found
that works is rebooting the machine -- which is very inconvenient because
of the number of users we have on our Risc-System 530 running long jobs.

So, 1. Has anyone else run into this problem (with automount and NFS mounts
       on non-IBM machines?)

    2. We're running 8 nfsd's and 15 biod's -- /etc/auto.direct has about
       15 entries.  Does that sound right?  Do you have to have a biod for
       each entry in /etc/auto.direct?

    3. Can anyone explain what is happening to us when this occurs?

    4. Is there a way to recover from this state without rebooting the machine?

    5. And the biggie:  Is there a way to *prevent* this problem?

    6. Sun's automount doesn't do this with exactly the same configuration.
       Should we just forget about using IBM's implementation?  Is there a
       public domain implementation out there (GNU automount?)?

Any help from an NFS-expert or other knowledgeable person would be greatly
appreciated by dozens of people around here.  Thanks in advance.  --Tim