31 Oct
2010
31 Oct
'10
4 a.m.
So I have these 2 PDP-8/L core stacks I am trying to recover:
One would be perfect, if bit 3 @ adr 0 would be alive...
Although this is 99.99% OK, it is of course not good enough.
The other stack seems to be a total loss :
2 sense wires are open circuit, more than 20 select diodes shorted. Of course the further
quality of the cores is unknown.
Is there any realistic way of getting one fully functional stack out of these ?
Removing a single core from the really bad stack to the almost OK stack would seems almost
feasible to me, since address 0 is bound to be on a edge of a core mat.
Any ideas / suggestions ?
Jos
31 Oct
31 Oct
6:47 a.m.
Removing a single core from the really bad stack to
the almost OK stack
would seems almost feasible to me, since address 0 is bound to be on a edge
of a core mat.
Any ideas / suggestions ?
Yes, core memory can be repaired by replacing individual cores and
restringing wires. Most (nearly all I have seen) stacks actually have
repairs if you look for them, as the production process never was
perfected. The manufacturers had women at benches reworking core
planes all day (what a job, eh?).
For your core plane, a rather common type, it may not be worth it.
Just wait until you can get a replacement, then put the bad plane up
on Ebay to pay for it.
--
Will
11:22 a.m.
On Sun, 31 Oct 2010, Jos Dreesen wrote:
So I have these 2 PDP-8/L core stacks I am trying to
recover:
One would be perfect, if bit 3 @ adr 0 would be alive...
Although this is 99.99% OK, it is of course not good enough.
Just to mke sure I'm hearing you right: Are you saying one individual core
is 100% dead, and everything else is 100% okay? If so - if it's one
single, individual core which is broken - I'd suspect this is pretty much
irrepairable. The options would be to A) try to form/mold/create a new
ferrite core around the wires, which would be insanely difficult assuming
it could be done at all, or B) to unwire that plane of the core stack,
replace the broken ferrite core with a new one, and then rethread/rewire
it. That's extremely nontrivial work, I have to believe. But they were
originally threaded by hand, albeit by people with microscopes and _very_
good eye-hand coordination and sewing skills, so it's within the realm of
possibility.
If I've read you wrong, though, and the whole thing is working except
address 0 bit 3 is working _some_ of the time, that would be really good
news. :) You could try scoping out the sense amps and inhibit drivers
for bit 3 and see if they were just borderline to specifications,
something which might be just tweaky enough to affect one bit.
The other stack seems to be a total loss :
2 sense wires are open circuit, more than 20 select diodes shorted. Of
course the further quality of the cores is unknown.
Oddly enough, this one might be easier to deal with. Have you tracked down
where the sense wires go open-circuit? If they're failing at a
solder-joint near an outer edge, you could potentially repair them without
having to open up the whole damn stack. The diode replacement would likely
mean partially disassembling the stack assembly to gain access to both
sides of the diode PCB, but at least you're only having to dig in one
level.
O'course making those repairs might simply reveal more problems at the
core level. But you've gotta start somewhere, I s'pose. :)
Is there any realistic way of getting one fully
functional stack out of these
?
Removing a single core from the really bad stack to the almost OK stack would
seems almost feasible to me, since address 0 is bound to be on a edge of a
core mat.
I suspect I'm not the only one on the list who:
-thinks opening up a core-stack and repairing it is
theoretically possible;
-also thinks it would be a hell of a daunting project;
-is somewhat amazed at the dexterity and patience of the people
who originally hand-wired them at manufacture; and
-thinks pulling off a repair of a core-plane by rewiring it by
hand would give significant bragging rights. :)
-O.-
11:51 a.m.
On Sun, 31 Oct 2010, O. Sharp wrote:
Date: Sun, 31 Oct 2010 14:22:35 -0400 (EDT)
From: O. Sharp <ohh at panix.com>
Reply-To: "General Discussion: On-Topic and Off-Topic Posts"
<cctalk at classiccmp.org>
To: "General Discussion: On-Topic and Off-Topic Posts" <cctalk at
classiccmp.org>
Subject: Re: Repairing core memories....
On Sun, 31 Oct 2010, Jos Dreesen wrote:
If you look at core planes, many have repairs visable in even in "new" planes,
probably ones that failed margin tests. I looks like the way cores in the
middle of the mat are replaced is by snipping a bit of the drive and sense
wires a couple of cores away from the faulty core, replacing the core and
threading in a bit of new wire for the drive and sense lines, with enough
slack to solder the splices above the plane. You can see the little splices in
many core planes (usually covered with a tiny drop of glytol or some such)
Peter Wallace
So I have these 2 PDP-8/L core stacks I am trying
to recover:
One would be perfect, if bit 3 @ adr 0 would be alive...
Although this is 99.99% OK, it is of course not good enough.
Just to mke sure I'm hearing you right: Are you saying one individual core is
100% dead, and everything else is 100% okay? If so - if it's one single,
individual core which is broken - I'd suspect this is pretty much
irrepairable. The options would be to A) try to form/mold/create a new
ferrite core around the wires, which would be insanely difficult assuming it
could be done at all, or B) to unwire that plane of the core stack, replace
the broken ferrite core with a new one, and then rethread/rewire it. That's
extremely nontrivial work, I have to believe. But they were originally
threaded by hand, albeit by people with microscopes and _very_ good eye-hand
coordination and sewing skills, so it's within the realm of possibility.
If I've read you wrong, though, and the whole thing is working except address
0 bit 3 is working _some_ of the time, that would be really good news. :)
You could try scoping out the sense amps and inhibit drivers for bit 3 and
see if they were just borderline to specifications, something which might be
just tweaky enough to affect one bit.
The other stack seems to be a total loss :
2 sense wires are open circuit, more than 20 select diodes shorted. Of
course the further quality of the cores is unknown.
Oddly enough, this one might be easier to deal with. Have you tracked down
where the sense wires go open-circuit? If they're failing at a solder-joint
near an outer edge, you could potentially repair them without having to open
up the whole damn stack. The diode replacement would likely mean partially
disassembling the stack assembly to gain access to both sides of the diode
PCB, but at least you're only having to dig in one level.
O'course making those repairs might simply reveal more problems at the core
level. But you've gotta start somewhere, I s'pose. :)
Is there any realistic way of getting one fully
functional stack out of
these ?
Removing a single core from the really bad stack to the almost OK stack
would seems almost feasible to me, since address 0 is bound to be on a edge
of a core mat.
I suspect I'm not the only one on the list who:
-thinks opening up a core-stack and repairing it is
theoretically possible;
-also thinks it would be a hell of a daunting project;
-is somewhat amazed at the dexterity and patience of the people
who originally hand-wired them at manufacture; and
-thinks pulling off a repair of a core-plane by rewiring it by
hand would give significant bragging rights. :)
-O.-
12:12 p.m.
I have a vivid memory of an IBM CE holding a broomstick with a magnet
on the end of it fishing out bits of fractured core out of the oil-
immersed memory of a 7090... :)
"Check the oil" was part of routine maintenance on that old iron.
--Chuck
4:12 p.m.
On Sun, Oct 31, 2010 at 3:12 PM, Chuck Guzis <cclist at sydex.com> wrote:
I have a vivid memory of an IBM CE holding a
broomstick with a magnet
on the end of it fishing out bits of fractured core out of the oil-
immersed memory of a 7090... :)
"Check the oil" was part of routine maintenance on that old iron.
--Chuck
So how does a single core fail? A crack I guess. Maybe overheating that
could change the characteristics of the core materials.
-chuck
4:44 p.m.
On 31 Oct 2010 at 19:12, Charles Dickman wrote:
So how does a single core fail? A crack I guess. Maybe
overheating
that could change the characteristics of the core materials.
That was my understanding--and the reason that core was in an oil
bath to start with.
I vaguely recall something related about the CDC 7600, where it was
possible to write pathological code that would repeatedly hit the
same bank of core (core, even on the PPUs was interleaved, but it was
very fast--for the time--27.5nsec core) and cause said core to
overheat and the machine to throw parity errors. The solution was to
integrate the frequency of bank accesses and slow the system down if
they became too frequent. I don't recall how it was done, however.
--Chuck
5:39 p.m.
Microdata core stacks had these huge resistors in the inhibit circuit
that would literally set the board and system on fire.
Panel mode, R register 0xA010 in the switchers (write) and run, wait for
smoke or PS failure.
Microcoded machines with easy access to this sort of 1/2 cycle write
would hit the board every 200us, when the requirement was for 600us for
a 1/2 cycle read, and 1000us for a Write. The writes would not complete
that fast, but the 1600 (and 800's) logic was such that you would cycle
the board fast enough to cause the damage.
I know that individual cores would simply stop functioning over time. I
worked with a company in Santa Ana, Memtec, which repaired the Microdata
and a few other local minicomputer companies core stacks, but mostly
Microdata's because the principles in the company were ex Microdata
field service.
I recall seeing one with the stack open and there was a small spot in
the middle of the stack where they had clipped out the failing core from
the three wires, extended (or rethreaded) the replacement back into the
stack, and it stood maybe 1/2mm to 1/4mm out from the array. They had a
dab of hotmelt or something on the rework since there were uninsulated
welds where they did the splicing. (mind you very small welds).
A friend acquired their working microscope in the early 80's for a
keepsake, and it was a large Nikon with binocular high eye relief eyepieces.
Microdata had two people who did corestack repair, and some of the
stacks made in Puerto Rico came with repairs, especially in the 16k
arrays. Microdata manufactured their own 16k core stack, but bought
outside for the 8K's. I think the original 4k stacks they used were
made by Dataram, I could look if anyone is interested.
The 4k stacks were probably dated from 1968 or so, and had a large metal
plate over the array which was about 8" by 8".
anyone have any idea about whether this was mechanical, or was it
possibly for magnetic reasons in early core stacks?
The 8K arrays and 16K arrays (as well as the Ampex / CDC arrays) were
all just circuit boards with a grid etch pattern covering them no full
metal plate.
Jim
On 10/31/2010 4:44 PM, Chuck Guzis wrote:
On 31 Oct 2010 at 19:12, Charles Dickman wrote:
So how does a single core fail? A crack I guess.
Maybe overheating
that could change the characteristics of the core materials.
That was my
understanding--and the reason that core was in an oil
bath to start with.
I vaguely recall something related about the CDC 7600, where it was
possible to write pathological code that would repeatedly hit the
same bank of core (core, even on the PPUs was interleaved, but it was
very fast--for the time--27.5nsec core) and cause said core to
overheat and the machine to throw parity errors. The solution was to
integrate the frequency of bank accesses and slow the system down if
they became too frequent. I don't recall how it was done, however.
--Chuck
1 Nov
1 Nov
7:21 a.m.
Date: Sun, 31 Oct 2010 17:39:14 -0700
From: jws at jwsss.com
To:
Subject: Re: Repairing core memories....
Microdata core stacks had these huge resistors in the inhibit circuit
that would literally set the board and system on fire.
Panel mode, R register 0xA010 in the switchers (write) and run, wait for
smoke or PS failure.
Microcoded machines with easy access to this sort of 1/2 cycle write
would hit the board every 200us, when the requirement was for 600us for
a 1/2 cycle read, and 1000us for a Write. The writes would not complete
that fast, but the 1600 (and 800's) logic was such that you would cycle
the board fast enough to cause the damage.
Hi
YOu mean ns, not us. 1000us is one millisecond. That is not
too fast.
I wonder what the failure mode is. It would almost have to be
that the cores cracked from thermal stress. I can't imagine
the magnetic properties changing otherwise.
A tiny air gap would make the cores hard to magnetize.
Dwight
10:11 a.m.
<snip>
Hi
YOu mean ns, not us. 1000us is one millisecond. That is not
too fast.
yep
I wonder what the failure mode is. It would almost
have to be
that the cores cracked from thermal stress.
the cores would nor fail, the inhibit
circuit's load resistors would set
the phenolic on fire.
I can't imagine
the magnetic properties changing otherwise.
A tiny air gap would make the cores hard to magnetize.
Dwight
31 Oct
31 Oct
7:29 p.m.
From: pcw at mesanet.com
---snip---
11:28 a.m.
So I have these 2 PDP-8/L core stacks I am trying to recover:
One would be perfect, if bit 3 @ adr 0 would be alive...
Although this is 99.99% OK, it is of course not good enough.
The other stack seems to be a total loss :
2 sense wires are open circuit, more than 20 select diodes shorted.
Of course the further quality of the cores is unknown.
Is there any realistic way of getting one fully functional stack out
of these ?
Removing a single core from the really bad stack to the almost OK
stack would seems almost feasible to me, since address 0 is bound to be
L on a edge of a core mat.
Its likely to be, I guess, but I wouldn't bet much on the physuical
layout of the core plane. In any case the sense and inhibit lines are
going to run through a lot of cores, so you'll have to remove and replace
those I think.
I have never tried to repari core memory (otehr than the drive
electronics), and I believe that fore some later core memoryies, the cors
were cast around the wiees after the wires had been laid out, so you
can't untrhead and retrhead that type. But older ones were made by
threading wires trhoug hthe cores, so it may be possible.
What I would do is practice on the totally dead plane. See if you can
remove the wires, take a core out, and then thread the wires back again.
When you get good at doing that, then have a go at fixing the almost-good
one
-tony
1 Nov
1 Nov
3:25 p.m.
On 2010 Oct 31, at 3:00 AM, Jos Dreesen wrote:
So I have these 2 PDP-8/L core stacks I am trying to
recover:
One would be perfect, if bit 3 @ adr 0 would be alive...
Although this is 99.99% OK, it is of course not good enough.
The other stack seems to be a total loss :
2 sense wires are open circuit, more than 20 select diodes shorted.
Of course the further quality of the cores is unknown.
Is there any realistic way of getting one fully functional stack out
of these ?
Removing a single core from the really bad stack to the almost OK
stack would seems almost feasible to me, since address 0 is bound to
be on a edge of a core mat.
Any ideas / suggestions ?
Here's a photo of some core with spliced wires:
http://www3.telus.net/~bhilpert/tmp/coreMontage.jpg
I suspect these were fixes done at the time of manufacture, there are a
quite a few of them over a large ~64KB planar array. I don't think they
were (all) done to replace cores though, they don't seem to open all
the wires to be able to replace a core. Perhaps fixing wires that broke
during manufacture.
Don't know how the electrical contact was made in the splices, whether
soldering or perhaps some sort of pressure weld.
I don't know what size cores you're dealing with, these ones are pretty
tiny, about 1/100 inch OD in 128*256 arrays.
Also included is an inset showing what happens when one drops a
screwdriver onto the cores (guilty). The plane is just a display item
though, so it wasn't a crisis.
This is a 3-wire system, there is no diagonally-strung sense-wire.
4947
days inactive
4948
days old
12 comments
10 participants
participants (10)
-
ard＠p850ug1.demon.co.uk -
cclist＠sydex.com -
chd＠chdickman.com -
dkelvey＠hotmail.com -
hilpert＠cs.ubc.ca -
jdr_use＠bluewin.ch -
jws＠jwsss.com -
ohh＠panix.com -
pcw＠mesanet.com -
wdonzelli＠gmail.com