Re: [Help-glpk] [Fwd: glpsol not converging]

[glpk xypron]

Hello Andrew,

looking at the output Narendra provided GLPK is stuck in the warm up phase and 
cannot find an initial LP solution. Here the time limit of the mip solver is 
ignored.

Would it be possible to pass the tmlim parameter to the simplex solver?

@Narendra
Looking at the value of your objective function "obj =   3.241402100e+07" could 
it be that you have very large ratios between the smallest and largest matrix 
elements of your problem?

Please, start your model again without --check and provide start of the output, 
e.g.
Model has been successfully generated
GLPK Integer Optimizer, v4.47
289 rows, 480 columns, 1680 non-zeros
240 integer variables, all of which are binary
Preprocessing...
288 rows, 480 columns, 1440 non-zeros
240 integer variables, all of which are binary
Scaling...
 A: min|aij| =  1.000e+00  max|aij| =  1.500e+01  ratio =  1.500e+01
GM: min|aij| =  9.586e-01  max|aij| =  1.043e+00  ratio =  1.088e+00
EQ: min|aij| =  9.189e-01  max|aij| =  1.000e+00  ratio =  1.088e+00
2N: min|aij| =  9.375e-01  max|aij| =  1.000e+00  ratio =  1.067e+00
Constructing initial basis...
Size of triangular part = 286
Solving LP relaxation...
GLPK Simplex Optimizer, v4.47

The ratio provided in the scaling phase gives a good indication if your model 
is well conditioned.

Analyze your model whether you are using some big M approach or dummy costs 
with very high numbers and try to find a better scaled formulation.

Best regards

Xypron


-------- Original-Nachricht --------
> Datum: Tue, 31 Jul 2012 16:41:40 -0700
> Betreff: Re: [Help-glpk] [Fwd: glpsol not converging]

> Hi Xypron,
> 
> Yes, I have re-checked the parameters again. I used 10000 (10E5) and not
> 10E6 for --tmlim.
> 
> The last 10 lines of output from the run are:
> 
> |16587000: obj =   3.241402100e+07  infeas =  1.037e-09 (2)
> |16587500: obj =   3.241402100e+07  infeas =  7.363e-10 (2)
> |16588000: obj =   3.241402100e+07  infeas =  5.294e-10 (2)
> |16588500: obj =   3.241402100e+07  infeas =  6.339e-10 (2)
> |16589000: obj =   3.241402100e+07  infeas =  9.333e-10 (2)
> |16589500: obj =   3.241402100e+07  infeas =  1.177e-09 (2)
> |16590000: obj =   3.241402100e+07  infeas =  3.791e-09 (2)
> |16590500: obj =   3.241402100e+07  infeas =  1.787e-09 (2)
> |16591000: obj =   3.241402100e+07  infeas =  8.604e-10 (2)
> |16591500: obj =   3.241402100e+07  infeas =  1.154e-09 (2)
> 
> This process is running on a 64-bit Linux m/c with 2G of memory reserved
> for this process. I also know that there is no slowdown happening because
> of any memory swapping etc. I am using v4.47 version as indicated by this
> message in the output of the run:
> 
> GLPSOL: GLPK LP/MIP Solver, v4.47
> 
> I also ran with --check option and the tool output is as below:
> 
> GLPSOL: GLPK LP/MIP Solver, v4.47
> Parameter(s) specified in the command line:
>  -m tam.mod -d step1.dat
>  --check
> Reading model section from tam.mod...
> 226 lines were read
> Reading data section from step1.dat...
> 2390 lines were read
> Generating IOSingleMode...
> Generating TotalInputIOs...
> Generating TotalOutputIOs...
> Generating IOPreInputMode...
> Generating IOPreOutputMode...
> Generating PreDefInput...
> Generating PreDefOutput...
> Generating UsedIOIsInput1...
> Generating UsedIOIsInput2...
> Generating UsedIOIsOutput1...
> Generating UsedIOIsOutput2...
> Generating InputIOUsedOnce...
> Generating OutputIOUsedOnce...
> Generating BlockInputIOUsedOnce...
> Generating BlockOutputIOUsedOnce...
> Generating FlatPreDefInput...
> Generating FlatPreDefOutput...
> Generating BlockInputFlatUnused...
> Generating BlockOutputFlatUnused...
> Generating FlatUsedIOIsInput...
> Generating FlatUsedIOIsOutput...
> Generating FlatBlockInputAssigned...
> Generating FlatBlockOutputAssigned...
> Generating FlatIOUsedOnceIn...
> Generating FlatIOUsedOnceOut...
> Generating FlatWireLengthIn...
> Generating FlatWireLengthOut...
> Generating WireLengthOut...
> Generating WireLengthIn...
> Generating WireLength...
> Generating Routing...
> Model has been successfully generated
> 
> I specified the --first option as I was trying different things to check
> if
> it helps in speeding up. I have also tried without that option, but I
> still
> get stuck in an endless run.
> 
> I really appreciate your help on this.
> 
> Thanks again and Best regards,
> -Narendra
> 
> 
> On Tue, Jul 31, 2012 at 4:22 PM, glpk xypron <address@hidden> wrote:
> 
> > Hello Narendra,
> >
> > if tmlim is 100000 (10E5) the optimization should have stopped after 28
> > hours. Please, check you did not enter 1000000 (10E6).
> >
> > On Linux you can use
> > ps -df
> > to check the parameters of your running process.
> >
> > To better understand your problem, please, provide:
> > - last output lines of your "endless" run
> > - operating system you run on (including whether its 32 or 64bit)
> > - version of GLPK you are using.
> > - output of glpsol -m tam.mod -d step1.dat --check
> >
> > The parameter --first typically will slow down the solution. You should
> > only specify it, if you are sure about how your variables are sorted and
> > why you want to always branch on the first integer variable.
> >
> > Best regards
> >
> > Xypron