17 Jul
2011
17 Jul
'11
4:42 p.m.
I have an Alphaserver 1000A which began intermittently halting, same
as if the halt switch had been pressed. At the time it had a serial
console. I thought it might have been picking up noise on the serial
line and interpreting it as a break signal so I changed over to a
graphics console. It still halted intermittently. This problem either
went away or the following problem caused me to not notice it any more.
The machine started intermittently powering itself down, requiring it
to be switched off and back on again to reset the power supply before
it could be run again. The machine might stay running for several
hours or it might power down less than a second after switching on, or
anything in between.
Thinking that the two problems might be related, I investigated the
operator control panel on which are mounted the low voltage on/off
switch and the halt switch. I found the two switches to have little
in common and my theory to be unlikely to be correct. The on/off
switch and associated circuitry appear to be working correctly, as
does the interlock which cuts power when the lid is removed.
I tried a power supply unit from another Alphaserver 1000A and found
this did not cure the problem so I replaced the original. Thinking
that the power supply might be sensing an overload, I disconnected
the power feeds to the disks/cd/floppy and removed all the PCI cards.
The problem was still there.
When it was failing regularly, I found a line going to the power
supply which normally sat at 0V and pulsed high just as the power
supply shut down. This line did not pulse the same way when the
power was turned off normally. I suspect it is a signal from the
system board to the power supply requesting it to shut down, or
less likely, a signal from the power supply to the system board
signalling it has a problem and is about to shut down.
I have a service guide which mentions the possibilty of the power
supply shutting down due to a fan failure or high temperature but
it doesn't give any details on how this works or even how to find
out if it is happening.
Does anyone have any detailed information about the power control
system in the Alphaserver 1000A or any helpful suggestions?
Regards,
Peter Coghlan.
18 Jul
18 Jul
1:45 p.m.
I tried a power supply unit from another Alphaserver
1000A and found
this did not cure the problem so I replaced the original. Thinking
that the power supply might be sensing an overload, I disconnected
the power feeds to the disks/cd/floppy and removed all the PCI cards.
The problem was still there.
When it was failing regularly, I found a line going to the power
supply which normally sat at 0V and pulsed high just as the power
supply shut down. This line did not pulse the same way when the
power was turned off normally. I suspect it is a signal from the
system board to the power supply requesting it to shut down, or
less likely, a signal from the power supply to the system board
signalling it has a problem and is about to shut down.
I have a service guide which mentions the possibilty of the power
supply shutting down due to a fan failure or high temperature but
it doesn't give any details on how this works or even how to find
out if it is happening.
That was my first thought (woithout reading the manuals...)
How many wires do the fans have? If they have 3 or more, it's quite
likely that they are power, ground, and soem kind of monitor line.
Possibly a frequency indicating how fast the fan is turning, possibly a
simple 'I#ve failed' signal. In any case I'd investigate there.
I assuem the fans are all running, and running somewhere near the right
speed?
Overtemperature might be a mechasnical switch, or a semiconductor
sensor (probably the latter).
Can you trace this shutdown sire from the PSU onto the mainboard and see
what it goes to? If you're lucky it'll be a recongisable chip (maybe a
microcontroller). See if that takes inputs from the fane or a temperature
sunsor.
-tony
5:10 p.m.
When it was failing regularly, I found a line
going to the power
supply which normally sat at 0V and pulsed high just as the power
supply shut down. This line did not pulse the same way when the
power was turned off normally. I suspect it is a signal from the
system board to the power supply requesting it to shut down, or
less likely, a signal from the power supply to the system board
signalling it has a problem and is about to shut down.
I have a service guide which mentions the possibilty of the power
supply shutting down due to a fan failure or high temperature but
it doesn't give any details on how this works or even how to find
out if it is happening.
That was my first thought (woithout reading the manuals...)
How many wires do the fans have? If they have 3 or more, it's quite
likely that they are power, ground, and soem kind of monitor line.
Possibly a frequency indicating how fast the fan is turning, possibly a
simple 'I#ve failed' signal. In any case I'd investigate there.
I assuem the fans are all running, and running somewhere near the right
speed?
Overtemperature might be a mechasnical switch, or a semiconductor
sensor (probably the latter).
Nothing stands out as obviously being a temperature sensor but I don't know
what one would look like. Maybe something in a metallic package, mounted
near the top of the vertically mounted system board? I can't see anything that
fits that bill nor anything on flying leads either.
Can you trace this shutdown sire from the PSU onto the mainboard and see
what it goes to? If you're lucky it'll be a recongisable chip (maybe a
microcontroller). See if that takes inputs from the fane or a temperature
sunsor.
Not really. It's a multilayer board and it is not at all obvious where the
trace goes. There are no likely suspects near the connector and there is lots
of other stuff such as the memory between the connector and any more likely
suspects. There is an LM339 and a largish 1 Ohm resistor over the other side
near where the fan connectors are so perhaps those are responsible for
monitoring the fans? There's nothing near those components that looks like a
likely temperature sensor.
The manual mentions a couple of jumpers which it says can be set to generate
a machine check rather than shutting off the power when there is a fan fault
or temperature fault. This looks worth a try now that I know the fan monitoring
at least could be responsible for the problem although I probably need to
install a different version of VMS first (one from before they messed up the
error log analyser. The version I have at the moment reports I have generated
some errors on a VAX 11/780 in 1996, 15 years before the creation date of the
error log file...)
Regards,
Peter Coghlan.
19 Jul
19 Jul
2:15 p.m.
There are two fans. Both run at what looks like a
reasonable speed. Both
have only two wires but they are plugged into three way connectors on the
system board. This led me to suspect that the manual was lying about the
fans being monitored (like it lies about the on-board video and a few other
items which are not fitted in my machine).
However, I found that if I unplugged or manually stalled either fan, the
power supply did indeed shut down about a second or two later, with the above
mentioned line to the power supply pulsing high like it does when the fault
happens. As there are no sensing wires to the fans, I guess the current drawn
by them must be monitored to generate the fault signal.
Ah, that has confirmed one thing, I think. That the signal in question is
some kind of safety shutdown. Now to find out what's triggering it.
On many occasions, the power supply has tripped out less than a second after
switching on. In that time, there is a brief flash from the power on LED and
the fans rotate somewhat. This is definately less time than it takes for the
circuit to react to me disabling the fans manually and is not enough time for
the fans to get fully up to speed either. So, I suspect the fans themselves
are ok and either the fan monitoring circuit or temperature monitoring circuit
is more likely to be faulty. (There is good airflow and nothing appears to me
to be getting too hot.)
OK...
Overtemperature might be a mechasnical switch, or a semiconductor
sensor (probably the latter).
Nothing stands out as obviously being a temperature sensor but I don't know
what one would look like. Maybe something in a metallic package, mounted near the top of the vertically mounted system board? I
can't see anything that
fits that bill nor anything on flying leads either.
I would use an ohmeter or a good continutiy tester to trace it. If
necessarty, stroke one probe along every pin on the board...
Can you trace this shutdown sire from the PSU onto the mainboard and see
what it goes to? If you're lucky it'll be a recongisable chip (maybe a
microcontroller). See if that takes inputs from the fane or a temperature
sunsor.
Not really. It's a multilayer board and it is not at all obvious where the
trace goes. There are no likely suspects near the connector and there is lots of other stuff such as the memory between the
connector and any more likely
suspects. There is an LM339 and a largish 1 Ohm resistor over the other side
near where the fan connectors are so perhaps those are responsible for
monitoring the fans? There's nothing near those components that looks like a
Quite possibly. Monitorign the current by the drop acrtoss the 1Ohm resistor.
I think I'd stick a a logic probe on the output of each section of that
339 in turn and see if you get any activity when you unplug or stall a
fan. Of coruse we dont yet know that he genuine shutdowns are caused by a
fan problem. But I guess you could also montiro the outpus fo the 339
when a real shut-down occurs.
likely temperature sensor.
The manual mentions a couple of jumpers which it says can be set to generate
a machine check rather than shutting off the power when there is a fan fault
or temperature fault. This looks worth a try now that I know the fan monitoring
I wonder what these jumpers are actually connected to...
Why do I wish there were scheamtics of this darn thing available...
-tony
21 Jul
21 Jul
5:29 p.m.
I think I'd stick a a logic probe on the output of each section of that
339 in turn and see if you get any activity when you unplug or stall a
fan. Of coruse we dont yet know that he genuine shutdowns are caused by a
fan problem. But I guess you could also montiro the outpus fo the 339
when a real shut-down occurs.
It's a surface mount chip with tiny pins and it's buried under a bunch of
wires going to the power supply. I think I'll try to explore other avenues
first.
likely temperature sensor.
The manual mentions a couple of jumpers which it says can be set to generate
a machine check rather than shutting off the power when there is a fan fault
or temperature fault. This looks worth a try now that I know the fan monitoring
I wonder what these jumpers are actually connected to...
Why do I wish there were scheamtics of this darn
thing available...
18 Jul
18 Jul
5:40 p.m.
At 03:42 PM 7/17/2011, you wrote:
I have an Alphaserver 1000A which began intermittently
halting, same
snip
When it was failing regularly, I found a line going to
the power
supply which normally sat at 0V and pulsed high just as the power
supply shut down. This line did not pulse the same way when the
power was turned off normally. I suspect it is a signal from the
system board to the power supply requesting it to shut down, or
less likely, a signal from the power supply to the system board
signalling it has a problem and is about to shut down.
That sounds like a crowbar circuit. If these are
linear power supplies, it's more likely to have a
crowbar, but I've seen some switchmode power
supply units with it. There's a simple circuit
somewhere that senses a failure that's in some
way dangerous to the equipment operated by that
power supply, maybe just a zener, resistor, and
maybe a capacitor too, to sense overvoltage.
Trips the crowbar by firing an SCR to short the
output so it will shut down abruptly. Called a
crowbar because it's like dropping a very large
conductor, like a crowbar, across the output terminals.
Since over voltage can fry a lot of expensive
kit, they use a crowbar to pop a fuse or breaker,
rather than risk more damage.
The crowbar is possibly adjustable, but I'd start
by checking the specs on output voltage and make
sure it's adjusted correctly on all output busses.
320 . [Philosophy] "I shall cheerfully bear the
reproach of having descended below the dignity of history." --Lord Macaulay.
NEW: a50mhzham at gmail.com ? N9QQB (amateur radio)
"HEY YOU" (loud shouting) ? Second Tops (Set Dancing) ? FIND ME ON FACEBOOK
43? 7' 17.2" N by 88? 6' 28.9" W ? Elevation 815' ? Grid Square
EN53wc
LAN/Telecom Analyst ? Open-source Dude ? Musician
? Registered Linux User 385531
19 Jul
19 Jul
10:45 a.m.
At 03:42 PM 7/17/2011, Tom wrote:
That sounds like a crowbar circuit. If these are
linear power supplies, it's more likely to have a
crowbar, but I've seen some switchmode power
supply units with it. There's a simple circuit
somewhere that senses a failure that's in some
way dangerous to the equipment operated by that
power supply, maybe just a zener, resistor, and
maybe a capacitor too, to sense overvoltage.
Trips the crowbar by firing an SCR to short the
output so it will shut down abruptly. Called a
crowbar because it's like dropping a very large
conductor, like a crowbar, across the output terminals.
I doubt a crowbar circuit is involved. A crowbar tends
to be used to blow a fuse or trip a mechanical circuit
breaker, not to trip an electronic overcurrent protection.
A single fault could both result in an overvoltage and a
failure to respond to an overcurrent condition. Activating
a crowbar circuit and relying on the overcurrent protection to
cut the power in that situation could result in a big mess.
Since over voltage can fry a lot of expensive
kit, they use a crowbar to pop a fuse or breaker,
rather than risk more damage.
Yes. However, there is no fuse blowing or breaker
being tripped in this case, just the power supply
shutting down.
The crowbar is possibly adjustable, but I'd start
by checking the specs on output voltage and make
sure it's adjusted correctly on all output busses.
I'm not sure what all the voltages are supposed to
be but the main ones appear to be correct. I've also
tried a substitute power supply with identical results.
It would be very unlikely that the two power supplies
are both producing overvoltage conditions intermittently.
Regards,
Peter Coghlan.
20 Jul
20 Jul
5:31 p.m.
I doubt a crowbar circuit is involved. A crowbar
tends
to be used to blow a fuse or trip a mechanical circuit
breaker, not to trip an electronic overcurrent protection.
A single fault could both result in an overvoltage and a
failure to respond to an overcurrent condition. Activating
a crowbar circuit and relying on the overcurrent protection to
cut the power in that situation could result in a big mess.
Actually, in my experience, a crowbar circuit is often used on a PSU
with electronic overcurrent protection, and said protectiuon should
operate if the crowbar fires due to an overvotlage. However, there's a
second line of overcurrent protection (e.g. a fuse) that will operate if
the electronic protection doesn;t (e.g. becasue the PSU control circuitry
is totally malfunctioning).
[There's a well-knwon PSU design known as a Boschert 2-stage. The basic
idea is that the recrified/smoothed mains (about 350V DC) is brought down
to 150V by a non-isolated swtiching regulator. This feeds a free-running
push-pull oscilaaotr driving the chopper transfortmer. The outputs from
that are rectified and feed the load. One otuput is fed back to the
control circuit of the 150V regualtor, and cuases its output votlage to
be adjusted so that the osiclator/chopper transformr produce the right
output votltages. There's an current sense circuit which monitors the
current drawn my the oscillator circuit which alos operates on the
control circuit of the 150V regulator. And a crowbar circuit on the
outputs which shorts one of thm to ground if an output voltage rises too
high. The extra current drrawn by the osciallotr in this condition should
then cause th e150V regualtotr to be shut down.
Now for th problem. There;s a common failure mode. It starts like this :
The chopper transitor in thr 15)V regualtor goes short circuit
The output of this regulator jumps to 350V.
All the ouptus of the PSU more than double as a result
The crowbar first, shorting an output to ground
This causes the overcurrent proteciton circuit to trip, removing the drive
from the 150V regualtor transistor.
Since that
transistor is shorted, remvoing the drvie doesn't do a darn thing.
The oscilator is working itno a shorted load. Bot its transistors short too
The load o nthe rectivfied mains is now the shorted transsitors, some
windings with a vry low DC resitance, and the current sense resistor. The
latter burns out
Mains now appears across a small-signal transitor which is blwon off the
PCB. The chopper control IC, other small components and even some PCB
tracks generally fail.]
>
>Since over voltage can fry a lot of expensive
>kit, they use a crowbar to pop a fuse or breaker,
>rather than risk more damage.
I'll add that even in the case I've jsut described, rebuilding the PSU is
a lot cheaper and less work than replacing every logic IC in the machine!
I'm not sure what all the voltages are supposed
to
be but the main ones appear to be correct. I've also
tried a substitute power supply with identical results.
It would be very unlikely that the two power supplies
are both producing overvoltage conditions intermittently.
Unless the PSU senses the voltage at the load to rliminate the votlage
drop in the wiring/connectors and this sense line is intermitant. So that
th PSU thinks it's not producing enough votlage nad is thus increasing
its ouput to try ot get the voltage o nthe sense wire high enough, and
then the crowbar trips.
Sensible PSUs have the sense lines conencted to the outputs via
resistors (around 100 Ohms) so that if th esense line does go
open-circuit, the output voltage doesn't rise too much.
-tony
9:06 p.m.
On 20 Jul 2011 at 22:31, Tony Duell wrote:
Actually, in my experience, a crowbar circuit is often
used on a PSU
with electronic overcurrent protection, and said protectiuon should
operate if the crowbar fires due to an overvotlage. However, there's a
second line of overcurrent protection (e.g. a fuse) that will operate
if the electronic protection doesn;t (e.g. becasue the PSU control
circuitry is totally malfunctioning).
I've got a couple of linear supplies that use just that design--an
SCR crowbar driven by a comparator following a LM723 regulator stage--
the 723 has overcurrent protection, so it all hangs together.
(and there is a fuse on the primary of the transformer as well)
--Chuck
21 Jul
21 Jul
2:24 p.m.
On 20 Jul 2011 at 22:31, Tony Duell wrote:
The HP9800 machines that I lkeep on going on about, and some of their
periphearlas are like that. They use 723s either as linear or switching
regulator controllers and have crowbars on the outputs. They do use the
current limiting circuitry of the 723, and if that doesn't help there's a
mains fuse.
-tony
Actually, in my experience, a crowbar circuit is
often used on a PSU
with electronic overcurrent protection, and said protectiuon should
operate if the crowbar fires due to an overvotlage. However, there's a
second line of overcurrent protection (e.g. a fuse) that will operate
if the electronic protection doesn;t (e.g. becasue the PSU control
circuitry is totally malfunctioning).
I've got a couple of linear supplies that use just that design--an
SCR crowbar driven by a comparator following a LM723 regulator stage--
the 723 has overcurrent protection, so it all hangs together.
(and there is a fuse on the primary of the transformer as well)
4750
days inactive
4754
days old
9 comments
4 participants
participants (4)
-
a50mhzham＠gmail.com -
ard＠p850ug1.demon.co.uk -
cclist＠sydex.com -
CCTECH＠beyondthepale.ie