[Clp] Clp trunk - windows problem

[Yann COLLETTE]

Hello,

Thanks.
I have noticed a big performance difference between linux and windows.
On the same problem, it takes 10 minutes under linux and 40 under 
Windows (on the same kind of machine).
With the same parameters (I use Clp has a library, not the executable), 
there are numerical differences (under windows, the pivot threshold is 
fined tuned by clp which is not the case under linux. I noticed this 
because on Cbc-2.8.9, there was an abort() in the pivot threshold fine 
tuning code which is not present anymore in the cbc / clp trunk).

What can explain this numerical behavior difference between windows and 
linux ?

I disabled the scaling under windows under clp and this disabling 
reduces the performance gap: 20 minutes instead of 40.

Best regards,

YC


Le 09/08/2014 13:17, Stefan Vigerske a écrit :
> Hi,
>
> ok, it's now using Sleep(1) on Windows, i.e. wait 1 millisecond.
> On other systems it's now usleep(100), which will wait 0.1 
> milliseconds - that should be the same as before.
>
> Stefan
>
> On 07/29/2014 01:56 PM, ycollette.nospam at free.fr wrote:
>> Hello,
>>
>> I tried to compile Cbc / Clp trunk under windows with the intel 
>> compiler and I found a problem: there is no nanosleep function under 
>> windows.
>>
>> I found an equivalent function for windows: Sleep which take a number 
>> of milliseconds as argument.
>>
>> #ifdef _WINDOWS
>> #include <windows.h>
>> #else
>> #include <unistd.h>
>> #define Sleep(x) usleep((x)*1000)
>> #endif
>>
>> Then use like this:
>>
>> Sleep(how_many_milliseconds);
>>
>> Best regards,
>>
>> YC
>> _______________________________________________
>> Clp mailing list
>> Clp at list.coin-or.org
>> http://list.coin-or.org/mailman/listinfo/clp
>>
>