[Help-glpk] solving large models with glpsol

[Andrew Makhorin]

SOLVING LARGE MODELS WITH GLPSOL

Solving large models may take a long time and require a lot of memory,
so in this case it may be more convenient to split the solution process
into three stages as described below.

1. Translate the model to GLPK format as follows:

      glpsol -m model.mod --wglp model.glp --check

   In this case glpsol reads the model from file 'model.mod', process
   it, and writes it to file 'model.glp' in GLPK format (which is
   a low-level text format similar to MPS format). The option '--check'
   tells glpsol not to solve the model right now.

2. Solve the model using the following command:

      glpsol --glp model.glp ... -w model.sol

   In this case glpsol reads the model in GLPK format from file
   'model.glp', solves it, and writes its final solution to a text file
   'model.sol'.

   On solving the model you may try to use different options, e.g.
   options to find an initial solution with a primal heuristic, options
   that specify the branching technique, options that enable generating
   cutting planes, etc.

   If the model is of MIP class, you may save to a file not only the
   final solution, but all intermediate integer feasible solutions
   discovered by the solver, with the option '--save filename'. The
   filename string specified in the '--save' option should contain
   an asterisk (*), which will be replaced by a file serial number.
   For example, if you specify '--save foo/bar*.txt', the first
   solution will be written to file 'foo/bar001.txt', the second to
   file 'foo/bar002.txt', etc.

3. To process the solution previously saved to a file you need to use
   the following command:

      glpsol -m model.mod -r model.sol [-o ...]

   In this case glpsol reads the model from file 'model.mod', process
   it, and reads its solution from file 'model.sol' as if the model
   were solved in a normal way. This, in particular, allows to execute
   model statements that follow the solve statement.