[Coin-discuss] Re: BCP eats up memory like crazy!
ladanyi at us.ibm.com
Wed Jan 4 22:35:07 EST 2006
BCP keeps the whole search tree. This means that if a cut has made it to the
"end" of a search tree node (i.e., when branching occurs) then the cut is
saved in the tree manager and there it lives forever. This would explain the
behavior you see.
I'll add a new method to the TM that will prune the deadwood in the search
tree. This method will be possible to invoke from the user code or it can be
called after every so many processed node.
At the moment I am busy with other things, but I should be able to get to this
before the end of this month. Feel free to bug me in private email if I don't
do it :-).
On Mon, 2 Jan 2006, Michael Hennebry wrote:
> On Mon, 2 Jan 2006, Miroslav Karamanov wrote:
> > Thank you, Michael. Limiting the density of the cuts fixes
> > the problem for the instances I solve.
> > I wonder how cuts are managed in BCP. What happens after a
> > cut is discarded from all subproblems where it has been
> > used? Is the cut removed from data structure or only
> > pointers to it are deleted?
> I've never used BCP, nor inspected its guts.
> Good memory management says to remove something
> before removing the last reference to it.
> Even with bad memory management it should no longer
> be affecting the size of the matrix factorizations.
> Mike hennebry at web.cs.ndsu.NoDak.edu
> May you live in interesting times. -- Drasnian blessing
> Coin-discuss mailing list
> Coin-discuss at list.coin-or.org
More information about the Coin-discuss