> I'm trying to solve a MIP with glpsol; it crashed, the last few lines in > the log were > > +2268089: mip = 9.717500000e+04 >= 9.128300000e+04 6.1% (271264; > 14832) > glp_malloc: memory limit exceeded > Error detected in file glpenv05.c at line 65 > > I quoted part of the log file below.
This is not a bug. Your mip instance is hard for glpk mip solver, and since the best bound selection technique is used by default, the solver failed due to insufficient memory. (Btw, --first is not a good choice for branching; try --drtom or --pcost.) > Also, but that's a different story, I wonder that the Cbc stand-alone > solver from the COIN-OR project could solve the same instance in less > than half a minute. I didn't expect the difference in running time to be > greater than a factor of perhaps 10 or 20, clearly not >600. Aren't both > using somewhat similar branch-and-cut techniques? Success/failure of the branch-and-cut method heavily depends on the cut generators used. In this sense coin-cbc is more powerful than glpk, especially for mip instances having non-binary variables like yours. _______________________________________________ Bug-glpk mailing list [email protected] http://lists.gnu.org/mailman/listinfo/bug-glpk
