[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Seg 11 in GCC

> I have done this
> upwards of 20 times (11 minute kernel compiles are fun, too) in the
> last 24 hours.  Never seen a SIG11.

OK - you're saying that a kernel VM problem caused those signal 11s.
They can also be caused by legitimate software bugs. However, there are
just too many reports (including one that I can duplicate on demand
using my own system, in front of you if you like) that _must_ be RAM
related because they go away when you change the RAM, and come back
when you put the old RAM back.

Note that it's not necessary for a dynamic RAM cell to fail (although
lots of them do indeed fail). It is much more often a memory read or
write timing problem. This is very likely to happen with consumer-grade
PC motherboards and RAM. They are designed to be cheap and run at the
very edge of the capabilities of the hardware. The speeds we are talking
about here mean that wires don't behave like wires anymore. They become
"transmission lines" and must be considered as chains of L-C circuits.
There is lots of room for timing problems because of this and other
details of high-speed but low-priced computers.

I'm speaking as an engineer who has written RAM diagnostics for image
computers (the Pixar I and II). I've verified that lots of them fail.
I thought it was especially funny to hear about the engineer who said
he'd only seen one RAM fail and had 35 years experience. I would have
to imagine that he started with vaccumm tubes, not RAM. 

People who work with commercial or military-grade equipment aren't very
familliar with this sort of problem, either. They have been paying for
the kind of designs that the consumer can't afford, and one of the
things they get out of that is strictly deterministic behavior from
their hardware, not the "just within tolerances" behavior common in
consumer PCs.

Pixar Animation Studios: Reality is not our business.
Pixar's "Toy Story" $184,592,498 domestic, $49 million overseas and counting.

Reply to: