[Couenne] Assessing performance on different systems

[Pietro Belotti]

Pete,

there are a number of factors at play here: the compiler version, the 
library version, the type of processor, the Linux kernel. It might be 
that the changes are partly due to other components used by Couenne 
also (Ipopt, Cbc, Clp, the ThirdParty packages).

I could only try your instance on two machines:

1) Linux 2.6.32, gcc 4.4.3, Intel(R) Core(TM) i7 CPU L 640
2) Linux 2.6.27, gcc 4.1.1, Intel(R) Xeon(R) CPU E5410

and got the same result, 5186 nodes, though less than the 8401 you got. 
If I can find a different result on other machines, I'll track down the 
reason (your example is small enough to make this task relatively easy).

Pietro


On 08/24/2010 12:15 AM, Pete Janes wrote:
> Dear all,
>
> I have been running the latest version of Couenne (using couenne, not
> couenne2) on different
> systems. The results are the same for all systems, as expected. But the
> number of nodes, the
> number of cuts, and the number of iterations required were different
> between each system. These
> differences make quite a big difference for larger problems.
>
> The systems used are as follows
> 1) Linux 2.6.13, gcc version 4.0.2, AMD Opteron(tm) Processor 240
> 2) Linux 2.6.28, gcc version 4.3.3, Intel(R) Core(TM)2 CPU 6400
> 3) Linux 2.6.18, gcc version 4.1.2, Intel(R) Xeon(R) CPU X3220
>
> Does any one know an obvious reason for these variations?
>
> I have attached a small test program, configuration file, and a printout
> of the results from system 2)
>
> Regards,
>
> Pete
>
> P.S I assume I didn't ask this question before in the past. It sounds
> familiar.
>
>
>
> _______________________________________________
> Couenne mailing list
> Couenne at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/couenne

-- 
Pietro Belotti
Dept. of Mathematical Sciences, Clemson University
Clemson SC 29634
Phone: 864-656-6765
Email: pbelott at clemson.edu
Web:   http://myweb.clemson.edu/~pbelott