analog failures

I just used some Microchip op amps--no complaints. And a quad i2c DAC. Nice stuff, actually.

AIUI you can clear nearly all the i2c deadlocks by having the master cycle the clock 9 times (or by just cycling the power :).

Cheers, James Arthur

Thing is, my clients typically do not want to hear the word "nearly" :-)

A vendor wanted to convince a client to adopt press-fit for motherboards. Something I never liked and where I have seen failures. Big meeting. Went like this:

"We guarantee 99.9% contact reliability on DIN connectors" ... a certain consultant raises hand ... "Sir, our motherboard has lots of DIN connectors and the pin number totals 3,456. According to your statement three to four of those could fail to connect reliably. Could you tell us which ones those would be?"

The meeting sort of ended.

Except those caused by some station clamping the clock low.

Ok, they do. Are we done?

Totally irrelevant to Schmitt triggers not glitching on noisy inputs.

That's what those engineers thought as well. Until the black eye at EMC. Hint: Non-Schmitt stages pass a lot of the noise towards their outputs and that can get radiated off. Or mostly conducted, picked up, and then radiated.

It's a design thing. If this works on a part, it will always work. (nearly ;-)

I was in one of those mettings when I worked in P'ok IBM. There was a problem that caused magic smoke to roll out of mainframes (customers were not happy). A VP demanded that after the fix it was impossible for this to happen again. Someone raised the point that never was a long time and nothing is perfect. The VP then said "OK, you can have _one_ failure, BUT IT HAD BETTER BE ON ONE OF OUR SYSTEMS!".

I remember you'd told that one before... ;)

They're used pretty often in server gear (at least, the economically-vintage hardware I've seen torn down: < 5 years age, 1/4" thick, 32 layer PCBs, you know, that sort of thing...). Those have at least as many pins, probably more inside the whole box.

Were they just a particularly s*it supplier, or what? :-)

Also DIN connectors, but nevermind that?

Tim

Oh Dear!

Press-Fit 5 row DINs are used in flight control systems on Boeing 777s and some Airbuses. It's a good job that they have a certain level of redundancy.

GH

This thread gives me deja vu. My keyboard, synthesizer, and Edirol DIN-5 to USB adapter all use DIN-5.

"Don't shoot me, I'm only the piano player."

Thank you,

I just designed a one-master system where that's not a danger--slaves can't drive the clock, it's just an input to them. It would certainly be bad if another master clamped your clock low, though.

I included a system-wide i2c reset for future cases. It cycles i2c power(!), which should handle even stubborn "clampers." :-0

Cheers, James Arthur

Which is why you'd want to design the thing so the i2c could never deadlock in the first place. But if you're like me, chastened by others' tales, you could include two levels of recovery so that even if the inconceivable happened it could ever be a critical fault, just in case.

Reliable + (redundancy + retry) = spandex, belt, and suspenders.

SPI can fail too if someone think's he's enabled when he isn't. What's the fallback then? But that doesn't much happen, right?

Cheers, James Arthur

he must have pulled that number from somewhere the sun doesn't shine, there is no way it is that bad

How do you just cycle power on just the I2C?

On Friday, February 24, 2017 at 5:20:55 PM UTC-8, snipped-for-privacy@notreal.com wrote: ...

The more recent SMbus devices are compatible with i2c but include a 25ms timeout to avoid system lockup even when clock-stretching.

kevin

Not really because the master enables and disables. So having a hung bus because or a rogue device isn't likely unless the device is physically damaged. Not so with I2C where I saw cases where a power cycle made it all come back and work fine. For a while ...

It was pretty bad, IMHO. One one line where they had a lot of field failures I convinced the client to send all motherboards over the solder bath. That kind of convincing was not easy, almost with kicking and screaming. But it sure fixed the problem.

We has some machines that had a lot more slots and probably close to

Those 96-pin ones in large card cages. That's what we always called them.

Eurocard/VME connectors, IOW.

Cheers

Phil Hobbs

We did often call the boxes VME boxes but for some reason never the connectors.

I still have two nice single-height Euro boxes for ham radio projects from decades ago. Never really used them, the way how it often goes.