[Coin-ipopt] Ipopt memory consumption

[Kirk Abbott]

Hello,

A few comments:
1) Never do the single precision thing
2) See if you can get ampl to build the problem
but not call ipopt. This will tell you how much
memory is being used by ampl
3) Your problem is reasonably dense...
I see about 10 nonzeroes/row in the Jacobian.
I am surprised about the number of nonzeroes in
the hessian. Consider a problem reformulation by
introducing intermediate variables. This can
help (see the papers on partial separability)
4) What is the largest problem that you expect to
solve? 50K vars, 100K vars? Do the simplest thing...
By more memory. Back in 1998, at one company we
had boxes with 1G RAM for solving some large
(200K variables, nonlinear problems). RAM is cheap.
My laptop and desktop all have 1G. Our servers
run closer to 6G.

I hope this helps.

Kirk.


--- coin-ipopt-request at list.coin-or.org wrote:

> Send Coin-ipopt mailing list submissions to
> 	coin-ipopt at list.coin-or.org
> 
> To subscribe or unsubscribe via the World Wide Web,
> visit
> 	http://list.coin-or.org/mailman/listinfo/coin-ipopt
> or, via email, send a message with subject or body
> 'help' to
> 	coin-ipopt-request at list.coin-or.org
> 
> You can reach the person managing the list at
> 	coin-ipopt-owner at list.coin-or.org
> 
> When replying, please edit your Subject line so it
> is more specific
> than "Re: Contents of Coin-ipopt digest..."
> 
> 
> Today's Topics:
> 
>    1. Ways to Trim Memory Usage? (Frank J.
> Iannarilli)
> 
> 
>
----------------------------------------------------------------------
> 
> Message: 1
> Date: Fri, 03 Jun 2005 21:03:06 -0400
> From: "Frank J. Iannarilli" <franki at aerodyne.com>
> Subject: [Coin-ipopt] Ways to Trim Memory Usage?
> To: coin-ipopt at list.coin-or.org
> Message-ID:
> <B7C946F12710C05CD434675E@[192.168.0.139]>
> Content-Type: text/plain; charset=us-ascii;
> format=flowed
> 
> Hi,
> 
> 
> I'm (successfully) running large application
> problems with IPOPT.  As I 
> increase their sizing, I'm looking for ways to trim
> IPOPT's memory 
> footprint.
> 
> Some vital statistics for my present memory
> performance:
> 
>    Win32 (compiled with MinGW), MA19/MA27, and using
> the AMPL interface;
>    Example problem -->> IPOPT itself eats 120MB
> memory:
>         "Going to allocate
>                double precision work space of size 
> 3355970
>                integer work space of size          
> 4320981
>          Number of variables           :    22869
>             of which are fixed         :        0
>          Number of constraints         :     9225
>          Number of lower bounds        :    22869
>          Number of upper bounds        :     4437
>          Number of nonzeros in Jacobian:   248913
>          Number of nonzeros in Hessian :   331776
> 
> 
> Some ideas -- I'm looking for comments:
> 
>    - Tried limited-memory BFGS (IQUASI=-6/6)  --->>
> Saves about 20MB or 20%
> 
>    - Is the 120MB memory consumption commensurate
> with the problem sizing 
> above?  Memory does not appear to be "leaking" over
> time.
> 
>    - Does the AMPL Subroutine Library (ASL) eat too
> much memory?  Would use 
> of the F77 interface make for substantial savings?
> 
>    - Use a "better/different" linear solver (e.g,
> more recent Harwell 
> routines, etc)?
> 
>    - Recompile to single-precision (is this even
> possible, not to mention 
> advisable)?
> 
>    - ?????
> 
> 
> Thanks!
> 
> 
> Frank J. Iannarilli, franki at aerodyne.com
> Aerodyne Research, Inc., 45 Manning Rd., Billerica,
> MA 01821 USA
> www.aerodyne.com/cosr/cosr.html
> 
> 
> ------------------------------
> 
> _______________________________________________
> Coin-ipopt mailing list
> Coin-ipopt at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/coin-ipopt
> 
> 
> End of Coin-ipopt Digest, Vol 7, Issue 1
> ****************************************
>