[Help-glpk] Write MIP model, read MIP model and then it's easy again!

[Pavel Klinov]

Hello all,

The question may seem weird but no more weird than the situation I'm observing.

Basically, my program calls GLPK to solve a series of MIP instances.
The instances are typically pretty easy and are solved within a
second. Sometimes I get some hard nuts that take seconds or dozens of
seconds but that's rare.

However, the program just stumbled across an instance that GLPK just
wouldn't solve in reasonable time. The first thing I tried was to dump
the model and run the cmd tool (glpsol). It solved it in seconds. Ok,
I thought that was because of the difference in some MIP control
parameters - but no. Then I tried to load that model from disk
*inside* my program and then call GLPK. Again, it solved it in
seconds.

That got me curious. Basically, the current situation is that GLPK
solves a number of MIP instances in a loop and eventually chokes at
one of them. But if you interrupt it, write the model to a file and
then read it back - it's all easy again!

Can it happen? It seems that depending on some internal state GLPK can
do some unfortunate move after which it just can't recover. But if you
write the model and read it back, the solver just won't go down that
hopeless path.

Can I dump some internal data structures to compare them before and
after the "write-read" operation? Is there anything else I can do
here?

I can provide the model and the terminal output(s) if it's of any help.

Thanks,
Pavel



-- 
cheers,
--pavel
http://www.cs.man.ac.uk/~klinovp