site map - email us - search  
 

General Questions

NCAR IBM Related Questions

Terrain

Regrid

Datagrid

Rawins

MM5

Graph

Conventions used in MM5

Trouble shooting and clues on various computers


 

MM5 Model on ALPHA

Running larger jobs on ALPHA

Many users have experienced problem running 'big' jobs on either single or multiple processor Alpha machines (big jobs mean that one either has large dimensions for MIX, MJX and MKX, or many nests). The typical error message is 'Segamentation fault', or

 
%DECthreads bugcheck (version V3.15-421), terminating execution.
% Reason:  Failure initializing the manager thread tcb (12)
% Running on OSF1 V4.0 on , 768Mb; 1 CPUs  


This is usually the indication that the job doesn't have enough memory to run. However this doesn't always mean that your machine doesn't have enough memory, most times, it means the machine isn't configured to run large jobs.

There are several steps you can try to 'fix' the problem.

  1. Type 'limit' on your machine and find out what you have. You should see something like

    cputime unlimited
    filesize unlimited
    datasize 1048576 kbytes
    stacksize 65536 kbytes
    coredumpsize 0 kbytes
    memoryuse 56284 kbytes
    vmemoryuse 1048576 kbytes
    descriptors 200

  2. If there are multiple processors on the machine, also try to set environment variable MP_STACK_SIZE by typing

    setenv MP_STACK_SIZE 96000000

  3. Then type 'unlimit' and 'limit' again to see what you have. You should probably see some changes in the numbers listed above. Many times, this should help, and you may start to run MM5 job.

  4. If you continue to have problem, and if you don't see the numbers change much before and after you type 'unlimit', and some numbers are small compared with the real memory you have on your machine (for example, if you have 1 Gb memory on your machine, and you don't see that displayed under the 'limit' you may need to reconfigure your machine.


MM5 input and output data on DEC ALPHA

Since DEC Alpha uses an IEEE data format (which is called lillte endian) that is different from most of worstations (which is referred to as big endian), we have included the following

-convert big_endian

in the compile options for DEC (see the configure.user file) in order for the model to use the same data as on other workstations. However, including this compile option implies that the output from MM5 on DEC is also in standard, big_endian IEEE format. Therefore if you want to run any application programs that read the model data, you would also need to include this compile option in your compilation.


Compliing MM5 on DEC ALPHA

The following patch is required to run MM5 model on DEC ALPHA with its new 4.1 Fortran compiler:

    *FORTRAN-OSF]FORTAUE01041

(provided by Digital)

UCAR Privacy Policy - Terms of Use - ©2003UCAR