Re: [eclipse-clp-users] LP solvers via eplex

From: Kish Shen <kisshen_at_...5...>
Date: Tue, 20 Jul 2010 14:33:37 +0100
Matthew Skala wrote:
> 
> That's not a trivial issue.  If you're doing the modelling in ECLiPSe and
> that requires a lot of memory, then there may not be much left for the
> external solver.  I ran into that with ILOG CPLEX.  A partial solution is
> to write an ECLiPSe program that will just call the external solver and do
> very little else, then run that as a separate process and have them
> communicate.  That's less convenient, but on a 32-bit machine it means you
> can have a basically separate 4G address space for the external solver.

Hi Matthew,

I am interested in hearing about user experiences with large problems 
with eplex -- how big a problem were you trying to solve? I am also 
interested in hearing about the relative amounts of memory required by 
ECLiPSe for the modelling, as compared to what the external solver need 
for solving.

By 'modelling', I assume you want to use the ECLiPSe model with other 
ECLiPSe solvers, such as ic? Otherwise you probably do not need the 
whole model in ECLiPSe at the same time and you can post the constraints 
incrementally to the external solver, and this is not a problem with 
CPLEX, but incrementally adding constraints using the COIN/OSI interface 
(as CBC/CLP does) is very slow (it looks like the whole problem is 
copied each time you incrementally add to it).

The memory requirements was discussed in our eplex paper (published in 
CP 2004), and the ECLiPSe representation of a problem uses around 4 to 5 
times more memory than the compact matrix used by the external solver.
However, it is clear that the solver need more memory to solve the 
problem, particularly if you are solving a MIP problem. The actual 
memory required depends on the nature of your problem of course, but I 
seem to remember there some MIP problems that were solved at IC-Parc 
(where  ECLiPSe was being developed) needed much more memory for solving 
than representing the problem, so that the 4-5 times more memory 
required by ECLiPSe to represent the problem was not so significant.

Cheers,

Kish

-- 
This e-mail may contain confidential and privileged material for the
sole use of the intended recipient. Any review, use, distribution or
disclosure by others is strictly prohibited. If you are not the intended
recipient (or authorized to receive for the recipient), please contact
the sender by reply e-mail and delete all copies of this message.
Cisco Systems Limited (Company Number: 02558939), is registered in
England and Wales with its registered office at 1 Callaghan Square,
Cardiff, South Glamorgan CF10 5BT.
Received on Tue Jul 20 2010 - 14:56:02 CEST

This archive was generated by hypermail 2.2.0 : Mon Jul 09 2018 - 02:05:29 CEST