[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Help-glpk] How to create a glp_prob pool?

From: Hruby Martin
Subject: Re: [Help-glpk] How to create a glp_prob pool?
Date: Sat, 15 Sep 2007 17:41:11 +0200 (CEST)


I solve a similar situation. It is also an extreme number of relativelly small LP problems.

I don't have such a good knowledge of the GLPK library from inside, so I cannot recommend any re-use of internal data structures. This could be also a source of possible problems.

I prefer solving the load of LP problems in parallel (in cooperation with boost-thread library) together with optimization of GLPK code in memory allocation. There is a plenty of "xcalloc" calls (memory alloc) which can be transfered to "static variables" on stack.

This optimized GLPK package will be hopefully available soon for testing.


On Sat, 15 Sep 2007, Ali Baharev wrote:

Dear Colleagues,

I have the following problem. In my computations i have to solve LP
problems repeatedly. The LP problems are not at all difficult, the
difficulty lies in the enormous number of LP problems to be solved.
One thing is common in all LP problems: they have the same form, i.e.
the non-zero elements of the constraint matrix always have the same
indices, the type of the rows and columns are always the same
(double-bounded or fixed). Here i neglected the fact that some entries
of the constraint matrix may become zero accidentally (if so, these LP
problems can be skipped as worst case solution).
I guess considerable time is spent on allocating and de-allocating
memory for the glp_prob objects, re-using a pre-allocated memory could
speed up the computations by orders of magnitude.
Is it possible to re-use the glp_prob object?
By re-use i mean replacing the whole constraint matrix, left and right
hand sides of the rows and columns. Does it make sense or the gain is
minimal if any?
Is there a way to minimize run-time memory allocation, and use a kind
of pool instead?

I would greatly appreciate any help,


Help-glpk mailing list

reply via email to

[Prev in Thread] Current Thread [Next in Thread]