kyw@cs.purdue.EDU (Ko-Yang Wang) (03/29/88)
In article <1461@ut-emx.UUCP> reeder@ut-emx.UUCP (William P. Reeder) writes: > >I have suggested to many sales reps that they need a bank of gee-whiz >lights on the front which blink faster as the load goes up. That way >people who walk by our machine room and look through the big window >could see something "impressive". The Pringle parallel computer (a 64 processor non-shared memory parallel computer that we designed at Purdue during 81-83 to emulate the CHiP architecture) does have several banks of red and green LED lights on it. (actually there are 128 of them, a pair of red and green LEDs for each PE) We programed the system so that a green LED lights up when the PE associated with it is doing I/O, and a red LED lights up when that PE is interrupted or finishes with its computation. When the Pringle starts to run, the red and green lights start to blink. I still remember when we successfully ran the first program on it, we turned of the lights in the room to enjoy the beautiful light show. (We repeated the execution several times just to see the lights.) Actually, we used the light pattern to detect deadlocks, when all the green lights stop blinking for longer than couple minutes and not all red lights light up - we knew we had a deadlock. This trick helped us to detect quite a few deadlock situations. >-- Wills >William {Wills,Card,Weekly,Virtual} Reeder reeder@emx.utexas.edu Ko-Yang Wang
eugene@pioneer.arpa (Eugene N. Miya) (03/30/88)
I saw the MIDAS multiprocessor at LBL some years ago solving a divide
and conquer problem (chess problem?). It was interesting to watch
the asynchronous processors each run similar light patterns then
converge on a solution.
Cray-2s have a set of LEDs on the top which you can read thru the transparent
fluorinert and window.
What was interesting about lights and sound was that the people working
on the MIDAS developed abilities to pattern recognize problems on the
basis of looking at the lights (as did numerous good PDP-11 programmers
and IBM light boxes of the 360 era). Not enough people can see a 2 to
develop this type of ability, but there might be CRI people who can.
A similar situation was shown to me while visiting Xerox PARC: an idle
loop bitmapped icon could tell some Dorado programmers what was wrong in
hardware. We are not a science yet, but it's interesting.
From the Rock of Ages Home for Retired Hackers:
--eugene miya, NASA Ames Research Center, eugene@ames-aurora.ARPA
"You trust the `reply' command with all those different mailers out there?"
"Send mail, avoid follow-ups. If enough, I'll summarize."
{uunet,hplabs,hao,ihnp4,decwrl,allegra,tektronix}!ames!aurora!eugenemarkd@rtech.UUCP (Mark P. Diamond) (03/30/88)
From article <3671@medusa.cs.purdue.edu>, by kyw@cs.purdue.EDU (Ko-Yang Wang): > In article <1461@ut-emx.UUCP> reeder@ut-emx.UUCP (William P. Reeder) writes: >> >>I have suggested to many sales reps that they need a bank of gee-whiz >>lights on the front which blink faster as the load goes up. That way >>people who walk by our machine room and look through the big window >>could see something "impressive". > Lights are more than just something to look at. On a multiprocessor system where each CPU has a light for when it is active the lights dynamically and instantaneously show system activity amoung all the processors. This is very useful in determining the amount of parallelism a program is achieving. Whenever I want to find out how an application is performing on our Sequent Symmetry I just go look at the lights. Mark <> <> <> <> Mark P. Diamond {sun, cbosgd, amdahl, mtxinu}!rtech!markd from Sequent Computer Systems onsite at Relational Technology