[fa.info-vax] FPS AP-120B Array Processor

info-vax (12/08/82)

>From ABOORNAZIAN@BBNG  Tue Dec  7 22:30:52 1982
Mail-From: ARPANET host BBNG rcvd at 6-Dec-82 1053-PST
Sender: ABOORNAZIAN@BBNG
To: info-vax@SRI-CSL
Cc: aboornazian@BBNG, depesa@BBNF, lanoue@BBNF, dipace@BBNF
Remailed-Date:  6 Dec 1982 2241-PST
Remailed-From: the tty of Geoffrey S. Goodfellow  <Geoff5 at SRI-CSL>
Remailed-To: Info-VAX@SRI-CSL: ;

We're having a problem with our AP-120B and I'd like to know
whether anyone else has seen something similar or has
suggestions.  Something in the interaction between the AP driver
and VMS during DMA transfers of results back from the AP to the
VAX causes the driver to get mixed up when the page it is trying
to write to is not in a working set.  One of two things happens
next.  Either the system crashes outright, or the AP hangs the
UNIBUS and it must be forcibly crashed (@CRASH).  This problem
appeared with the advent of VMS 3.0.  We are running the latest
version of the FPS software, the version which is supposed to
work with VMS 3.0.  Any clues would be greatly appreciated, and,
if you depend on an AP-120B and haven't gone to VMS 3.0, a word
to the wise is sufficient.

info-vax (12/10/82)

>From ron@BRL  Fri Dec 10 02:40:40 1982
Mail-From: ARPANET host BRL-BMD rcvd at 9-Dec-82 0924-PST
To: ABOORNAZIAN@Bbng
Cc: info-vax@Sri-Csl, aboornazian@Bbng, depesa@Bbnf, lanoue@Bbnf, dipace@Bbnf
Remailed-Date:  9 Dec 1982 2209-PST
Remailed-From: the tty of Geoffrey S. Goodfellow  <Geoff5 at SRI-CSL>
Remailed-To: Info-VAX@SRI-CSL: ;

We have an FPS AP120B which worked fine by itself on our 11/70 but
died miserably in a multiuser situation when other peripherals were
in use.  The symptoms were Unibus hangs and interference with other
devices.  We got no help from FPS (nick named Fly-By-Night Convolvers).
After talking to the guys at Stanford who wrote the UNIX software
for the thing, we discovered there was a design error in the UNIBUS
interface board (which FPS knew about but refused to give us an ECO
without us paying $900).  The problem is that the interface in the
interupt procedure places the vector on the address lines and then
before waiting the specified deskewing time, they send the INTR signal.
All kinds of hell break loose on the bus.  The fix was to run the INTR
signal through a couple of spare inverters on the side of the board.
Another problem is that we were never able to get the board to work
with the throttle count set above 1 or 2.  (The installer had set it
to 16 or something like that).  This is a little rotory switch at the
top of the UNIBUS interface board in the AP cabinet.

Hope this helps.

-Ron