Monday, December 14, 2009

GAMS/Mosek feedback

GAMS/Mosek seems to have some issues with a few (very non-critical) test problems: http://amsterdamoptimization.com/models/blog/mst.gms and http://amsterdamoptimization.com/models/blog/mst51.gms.

Some suggestions/observations:

  1. The listing file does not give a hint about why the run failed. I just see:

                   S O L V E      S U M M A R Y

         MODEL   m                   OBJECTIVE  z
         TYPE    RMIP                DIRECTION  MINIMIZE
         SOLVER  MOSEK               FROM LINE  134

    **** SOLVER STATUS     4 Terminated By Solver     
    **** MODEL STATUS      6 Intermediate Infeasible  
    **** OBJECTIVE VALUE                0.0000

    RESOURCE USAGE, LIMIT          5.053      1000.000
    ITERATION COUNT, LIMIT         0        100000

    MOSEK Link       Nov  1, 2009 23.3.2 WEX 13908.14598 WEI x86_64/MS Windows

    M O S E K       version 6.0.0.53 (Build date: 2009-11-11 07:16:11)
    Copyright (C)   MOSEK ApS, Fruebjergvej 3, Box 16
                     DK-2100 Copenhagen, Denmark
                    
    http://www.mosek.com

    It would be nice to see some error message about what happened.
  2. The last message on the screen is Return code - 1051 [MSK_RES_ERR_SPACE]. This message is not explained in the GAMS/Mosek documentation. I would prefer a message in readable English so I don't even have to check the manual. 
  3. I am guessing that the message has to do with running out of memory. However this is a 64 bit system and the models are not extremely large. So may be I stumbled on a bug here. I would guess that Mosek would normally be a top-contender on this kind of model.

PS. Later MOSEK versions solve these models ok (see comments).

PS 2: Unfortunately, I still see the same problems with:

               S O L V E      S U M M A R Y

     MODEL   m                   OBJECTIVE  z
     TYPE    RMIP                DIRECTION  MINIMIZE
     SOLVER  MOSEK               FROM LINE  133

**** SOLVER STATUS     4 Terminated By Solver     
**** MODEL STATUS      6 Intermediate Infeasible  
**** OBJECTIVE VALUE                0.0000

RESOURCE USAGE, LIMIT          5.272      1000.000
ITERATION COUNT, LIMIT         0        100000

MOSEK Link       Nov  1, 2009 23.3.3 WEX 14734.15043 WEI x86_64/MS Windows

M O S E K       version 6.0.0.55 (Build date: 2009-11-24 09:29:07)
Copyright (C)   MOSEK ApS, Fruebjergvej 3, Box 16
                 DK-2100 Copenhagen, Denmark
                
http://www.mosek.com

6 comments:

  1. I just ran the latest MOSEK 6.0.0.55 + GAMS on mst and it solved RMIP fine.

    The message means MOSEK ran out space. Nowadyas GAMS is repsonsible for link so they can certainly add better error messages if they like :-).

    I will talk to GAMS about this issue.

    ReplyDelete
  2. It is not really fixed. MOSEK does a very large allocation of 8GB so if you have a large computer as I then there is no problem. We are looking into if the 8GB allocation is really needed. Most likely not.

    In any case if you use the dual simplex optimizer rather the default interior-point optimizer will give you much lower storage consumption and much better performance in this case.

    ReplyDelete
  3. We have found a bug in the memory allocation so we in some cases allocate way too much space. I will update GAMS as soon as I have verified the bug fix does not have any unintended side effects.

    ReplyDelete
  4. I had the same problem.
    At first, I thought that there may be something wrong with my GAMS model...

    ReplyDelete