[Help-glpk] concatenating small LPs

[Andrew Makhorin]

>Is it interesting to concatenate a lot of (very)small-sized LPs into a big
>one for GLPK ?
>I'm thinking in terms of:
>       - memory usage
>       - speed of solving
>       - speed of setting/accessing the data
>       - precision

If a set of lp's are independent on each other, it is better to solve
them separately. Merging them into one instance would involve increasing
of overhead expenses, although it is possible to do that.

On the other hand, if lp's in a set are connected with each other by
means of some additional rows and/or columns, it again is better to
solve them separately, then to merge them into a big one including all
the additional rows and columns, and to solve the big lp using a union
of optimal bases of the lp's from the set as an initial basis for the
big lp. If it is possible to organize computations in such way, it may
be much more efficient than just to solve the big lp in an usual way.
The problem, unfortunately, is that such a decomposition is, as a rule,
a priori unknown.

Which case do you mean?