running FET on multicore processor

Started by catalin, April 08, 2009, 01:47:49 PM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

catalin

Hi,

during my testing of FET I've tried to run it on 3 different machines:
1- my laptop: Windows XP 32bit, Intel Centrino Duo, 2 Ghz, 2 GB RAM
2- server: Windows 2003 R2, 64 bit, Intel Xeon E 5345 2.33 Ghz (2 x quad core)
3- desktop: Windows XP 32bit, Intel dual core 2, 2 Ghz

the worst performance were on the server (2), executing "sample_inputs\Romania\Faculty-Econ-Timisoara-difficult\2007-2008-sem-1" in more than 2 hours

The same sample was executed on my laptop in ~50 min.

On both machines I've seen that the application uses only one core of the processor (in both Task Manager and Process Explorer from SysInternals)

I expected the sample to be executed faster on the server so my question is if it is possible to implement such improvements ? (e.g. Qt Concurrent or other similar)?

Thank you,
Catalin

Liviu Lalescu

#1
The time of generation is random.

If you want to compare, you need to set the random seed to the same value (in generate.cpp, add in Generate::generate(...) a "extern int XX" and set XX to the same value and recompile). Ask me for details if you don't succeed, I'll detail.

In the past, I did not know about multiple cores, so I didn't bother.

It might be possible to speed up single generation, but it seems very difficult.

It might be easy to speed up multiple generation, to make a single generation on each core. But I prefer the present stability of FET. Maybe in future versions I'll try.