[Gllug] Diagnosing hardware faults

[tid]

On 30 November 2010 20:11, Nix <nix at esperi.org.uk> wrote:
> On 28 Nov 2010, John Hearns stated:

> I still swear by a GCC bootstrap-and-test. Kernel compiles check for
> pointer-chasing in the compiler, but single-bit flips elsewhere which
> merely lead to bad code will not be caught. Because a GCC bootstrap
> builds the compiler with itself three times and tests the result, it
> tends to find problems more effectively. (I've had RAM problems in the
> past which only manifested themselves by a change in GCC test results
> between a good machine and a bad one. The bad one's results were even
> consistent between runs: just different from the good machine's.)

heh. Nice one. Had a similar test with memory errors when running distcc in
the past. Out of a farm of 20 identical Dell 1850s, two would routinely fail
when running a distributed build. It took a plethora* of Dell Engineers
to eventually diagnose the problem.

Tid

*Not sure what the collective noun for onsite engineers is - a coven?
a saddness?
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug