[Gllug] server sleeping

[Nix]

On Thu, 14 Sep 2006, Daniel P. Berrange gibbered uncontrollably:
> On Thu, Sep 14, 2006 at 01:02:42AM +0100, Nix wrote:
>> In my experience a good few GCC bootstraps can find memory and CPU
>> faults that a memtest86 can easily miss.
> 
> In large SMP machines that typically won't stress things enough because
> it won't utilize enough resources strenuously enough.

Multiple bootstraps with make -j {num-CPUs} bootstrap check are quite
likely to, even with a silly number of CPUs :)

>                                                       Last time I did stress
> tests I used a combination of things. Given a N logical cpu machine, used N
> copies of IOZone each doing IO in sizes > physical RAM (ensuring kernel VM 
> can't cache all the IO), N parallel kernel compiles (to ensure the schedular
> had lots & lots of processes to move around between CPUs), and N copies of 
> a program doing a memory bandwith test (to max out the entire memory sub-
> system). That could normally kill the faulty boxes in a couple of hours 
> with ECC errors, or just a spontaneous reboot :-)

Yeah. The reason I'd use a GCC bootstrap rather than a kernel compile in
that situation is simply because the GCC bootstrap *runs* the code that
gets built, so that even tiny intermittent problems are likely to get
picked up, as a miscompilation/comparison failure/testcase variation
if nothing else.

-- 
`In typical emacs fashion, it is both absurdly ornate and
 still not really what one wanted.' --- jdev
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug