The Joys of a Sun Hardware Upgrade, ois.db Style

From: John DiMarco 
Subject: The complete ois.db hardware upgrade saga

Here's a war story to rival anything from Panama. Jeff has been giving some
play-by-plays; here's the whole story.

One fine day, the DB powers-that-be had an idea: let's replace all the DB
diskless Sun 3/50s with diskless Sparcstations. Let's do it a couple
of machines at a time....

So they ordered a couple of diskless 4/60s. But then they realized that
ois was too slow to adequately fileserve a whole bunch of sparcstations. So
they had another idea: Let's upgrade ois.db to a 4/280. 

So they paid sun ~$14K and got a Sun 4 CPU board for ois.db. And it fell to
me to install it.

Some preliminary info: ois.db is a Sun 3/280 with 8 Mbytes of memory, 1
SCSI controller with a 1/4" tape drive, 1 Cyprico 3223 disk controller with
2 Swallows, and 1 Cyprico 3200 disk controller with 2 Swallows. We boot off
the 3223 in Xylogics emulation mode. 

I installed a Sun 4 filesystem using the backup root partition (on the 3223's
2nd swallow) and other partitions on the various disks, then Bernie and myself
(with help from Ken) tackled the upgrade. I set it up so that we could switch
back and forth from Sun4 to Sun3 operation by merely swapping the CPU boards
(fortunately we had the disk space). 

We replaced the Sun3 board with the Sun4 board... machine wouldn't boot
off the 3223. We could boot partially off the net (get BOOT from fmrl, vmunix
from disk), but ois would panic every couple of minutes. And the Sun4 board
failed its extended memory tests.

So we screamed and hollered at Sun. After giving them the revision number of
the Sun4 board (Revision 11), they told us that they didn't understand why
we had gotten such a low revision (according to ANT, Rev. 14 is the lowest
working Rev. for Sun4 CPU boards), and they promised to send us a new CPU board
and memory board... by overnight courier (growl!). We told them to forget the
courier; we'd send a taxi (we paid!). 

So we got CPU board #2 (and another memory board). This one was revision 24.
It fixed the PANIC problem, but we still couldn't boot off the 3223. This
3223 controller had worked correctly in ephemeral, so we figured we had another
bad CPU board. We complained to Sun again, suggesting they TRY the board 
before sending it to us.

By now, Sun decided that we were worth paying more attention to, so they
sent us CPU board #3, and we sent back board #2. And this time Sun paid the
courier. The new board took up two slots (rather than one), and was Rev. 6 (!).
This gave us a bit of a panic until Sun explained that their double form-factor
boards had a different revision numbering. 

We tried the new board... same problem as board #2. In the meanwhile, Sun
had tried the board we sent back (using a Xy451 controller), and it had worked
correctly. We dug up a Xy451 controller ourselves, and tried it with board #3,
and it worked too. Conclusion: the CPU board was probably ok. So we called
Cyprico. They had no idea what the problem might be, but knew of somebody at
the University of Utah who had the same problem.

So: we had a 3223 which worked in a Sun4 (fmrl) and a Sun4 CPU board which 
worked with an XY451, but the two wouldn't work together. What was left? 
The backplane and the power supply. But the power supply was fine.

So we asked Cyprico to try one of their controllers in a 3/280 with a
Sun4 CPU board. They promised to try it, and after weeks of waiting and lots
of calls, they finally did it. It worked. But their 3/280 was a much more
recent model than ours...

So we called Sun, and asked for a new backplane. They sent us one, and we
installed it. No difference. Grrr. We were out of options. ois.db just didn't
want to be a sun4 (do you blame it?).

Sun had done all they could at this point, so we decided to bother Cyprico
again. We suggested they replace our controller. They suggested they replace
the PROMS on our controller with their most recent revisions (rev. 3, I think.).
So they sent us new proms. We tried them yesterday. They worked! So it appears
the ois.db hardware upgrade saga is over, for now. Shortly we will switch to
the new board permanently, and ois.db will be a Sun4. 

						- John

Ha, ha, ha. Take me back to [ the alphabetic list ] [ the date-ordered list ].