[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

20030227: MUG Inquiry Change: 12021



>From:  address@hidden
>Organization:  SSEC
>Keywords:  200302272136.h1RLast03182 McIDAS GRID

Hi,

>We updated MAXGRIDPT to 400000 about a year ago in inquiry 11634.  One of
>the inquiry states that HP and SUN workstations may break at values higher
>than 400000.  If this is true, is there a possible workaround for John
>and his FL Exp?


HP is notorious for having a small stack size in the kernel, and this
limits the size of statically allocated arrays in programs.  The following
snippit from GEMPAK release documents tells the story for HP:

   *******  IMPORTANT PLEASE NOTE.  *******

   The standard kernel values for HPUX11 stack size MUST be increased
   to the following to account for the increased grid size:

           maxssiz =   134,217,728
           maxtsiz = 1,073,741,824
           maxdsiz = 1,073,741,824

   If these parameters are not increased, NMAP/NMAP2 core dumps at start
   up.

   Other systems tested at the Computing Development Branch (CDB) - which
   include HPUX10, LINUX, IRIX6, AIX4, SUNOS5 - did NOT require any kernel
   changes.

Sun doesn't have the same problem as far as I know.

The "real" solution to the problem is to be able to use arrays that
are created dynamically.  Unfortunately, this is not possible in Fortran
77, but it is possible in Fortran 90/95 and in C.

So, my recommendation would be to try to increase the array sizes on
Sun straight off and run some tests (JSFC has done this, on their
machines, but I don't know if they are using Sun or something else).
For HP, I would look into the kernel mods noted above.

>Inquiry #
>12021
>
>Summary
>High resolution grids are expected for experiment over Florida; MAXGRIDPT
>must be increased.
>
><a href="http://www.ssec.wisc.edu/cgi-bin/MUG/view/Show.cgi?FoundRecord=12021.
> dat&Bare=1">12021</a>

>From address@hidden Fri Feb 28 11:53:30 2003
>CC: address@hidden, address@hidden, address@hidden,
>   address@hidden, address@hidden,
>   address@hidden, address@hidden
>Subject: Re: 20030227: MUG Inquiry Change: 12021

The other thing to do, which is done in other programs with large 
arrays, is to put the arrays in common blocks because they don't end up 
in the stack [at least the way I understand it].

IMGREMAP has arrays of  one million words.

dave