[Help-glpk] MIP performance in last GLPK versions

[Prof. Hartwig Baumgaertel]

Folks, 

I have a question regarding performance of MIP solving with GLPK
versions 4.20 and 4.25 . 

I do some experiments with a MIP model for production location
optimization and observed 
that the performance went wrong from 4.20 to 4.25. 

I compiled both versions in the same way, on a WinXP PC with Cygwin
distribution. 
I have installed and compiled gmp 4.2.1. which is strongly recommended
during configuration 
of the distribution. 

With 4.20 my model with the reference data set was solved in 105 seconds
using 5 MB. 
4.25 needs 275 seconds and 69 MB memory: 

----- 4.25 --------
Model has been successfully generated
glp_simplex: original LP has 83 rows, 1722 columns, 5125 non-zeros
glp_simplex: presolved LP has 82 rows, 1722 columns, 3403 non-zeros
lpx_adv_basis: size of triangular part = 82
      0:   objval =   1.839035934e+06   infeas =   1.000000000e+00 (0)
     77:   objval =   2.016648319e+06   infeas =   0.000000000e+00 (0)
*    77:   objval =   2.016648319e+06   infeas =   0.000000000e+00 (0)
*   200:   objval =   9.100427277e+05   infeas =   9.100747288e-14 (0)
*   235:   objval =   8.918613118e+05   infeas =   1.421085472e-14 (0)
OPTIMAL SOLUTION FOUND
Integer optimization begins...
+   235: mip =     not found yet >=              -inf        (1; 0)
+   341: >>>>>   1.009709000e+06 >=   8.918613118e+05  11.7% (37; 0)
+  1212: >>>>>   1.003047000e+06 >=   9.195034779e+05   8.3% (340; 10)
+  1408: >>>>>   9.739900000e+05 >=   9.220886397e+05   5.3% (400; 35)
+  2493: >>>>>   9.731720000e+05 >=   9.301084986e+05   4.4% (678; 403)
..... 
+170839: mip =   9.644750000e+05 >=     tree is empty   0.0% (0; 70607)
INTEGER OPTIMAL SOLUTION FOUND
Time used:   275.0 secs
Memory used: 69.3 Mb (72666037 bytes)
--------- 4.25 end ---------------------

--------- 4.20 -------------------------
Model has been successfully generated
glp_simplex: original LP has 83 rows, 1722 columns, 5125 non-zeros
glp_simplex: presolved LP has 82 rows, 1722 columns, 3403 non-zeros
lpx_adv_basis: size of triangular part = 82
      0:   objval =   1.839035934e+06   infeas =   1.000000000e+00 (0)
     77:   objval =   2.016648319e+06   infeas =   0.000000000e+00 (0)
*    77:   objval =   2.016648319e+06   infeas =   0.000000000e+00 (0)
*   200:   objval =   9.100427277e+05   infeas =   9.100747288e-14 (0)
*   235:   objval =   8.918613118e+05   infeas =   1.421085472e-14 (0)
OPTIMAL SOLUTION FOUND
Integer optimization begins...
+   235: mip =     not found yet >=              -inf        (1; 0)
+   390: mip =   1.003727000e+06 >=   8.918613118e+05  11.1% (35; 0)
...
+161291: mip =   9.644750000e+05 >=   9.344634178e+05   3.1% (209;
48366)
+164265: mip =   9.644750000e+05 >=     tree is empty   0.0% (0; 50447)
INTEGER OPTIMAL SOLUTION FOUND
Time used:   106.0 secs
Memory used: 5.0 Mb (5289874 bytes)
---------- 4.20 end ----------------------

Can somebody explain me this difference?  (I solved the problem by
glpsol, 
without using any MIP specific parameters.) 

Thank you, 

Hartwig 

----
Prof. Dr.-Ing. Hartwig Baumgärtel 
Hochschule Ulm - University of applied sciences 
Institut für Betriebsorganisation und Logistik
Prittwitzstrasse 10
89075 Ulm 
Germany 
e-mail: address@hidden
Web : http://www.hs-ulm.de/baumgaertel 
Tel : +49 731 50 28 281
--