Re: [Boost-testing] Timeout on Martin Wille x86-64 machine?

From: Martin Wille (mw8329_at_[hidden])
Date: 2007-11-25 11:40:44

John Maddock wrote:
> Martin Wille wrote:
>> John Maddock wrote:
>>> Martin Wille wrote:
>>>>> Any chance of upping the limit slightly?
>>>> Yes. I'll set a higher limit for those three tests.
>> The three tests are "green" now.
> 
> Not quite see below...
> 
>>> Many thanks Martin: actually there's a forth long runner as well:
>>> test_binomial_real_concept.
>> This one hits a wall-clock-time limit now. Does it use a lot of
>> memory? I'm asking because the test didn't hit the 40s CPU time limit
>> and I suspect it caused swapping (bjam takes more than 2GB now, so
>> there's always something being swapped in and out) which would be
>> responsible for the delay.
> 
> Hmmm, I don't see the wall clock message in the current log: 
> http://tinyurl.com/32dj3t
> On my old-ish laptop this takes 46s to run, and uses 3Mb of memory.
> 
> Ah now wait, the gcc-4.2.1_linux_x86_64 results do display the "wall clock" 
> timeout message, but not the gcc-3.4.6_linux_x86_64 results.
> 
> Likewise the test_igamma_inva_real_concept has the wall clock timeout for 
> gcc-4.2.1_linux_x86_64  but not for gcc-3.4.6_linux_x86_64.
> 
> Again on my laptop, this one takes 37s to run and uses ~3Mb memory.

Well, your oldish laptop likely is more modern than the machine I use
for testing. I'll increase CPU time for the four tests again.

3MB isn't a lot, so swapping should not be an issue. However, I doubt
the test is I/O bound for any reasons other than possible swapping. I'm
clueless here. I'll increase the wall clock time limit, too.

HTH,
m