Thanks to all who responded so fast.
Thomas Leitner <tom_at_radar.tu-graz.ac.at>
"Hostrander, Natalia" <nhostrander_at_PMARE.com>
Matt Phelps <mphelps_at_cfa.harvard.edu>
Arno Hahma <arno_at_utu.fi>
Russel Morison <rzm_at_lighthill.maths.unsw.edu.au>
"Partin, Kevin S" <Kevin.Partin_at_SW.Boeing.com>
The (partly) solution to the problem was indeed to increase the
data segment size.
In principle was it my poor understanding of Unix.
Thomas was very fast and adviced me to the normal commands
csh> limit datasize unlimited
csh> matlab
or
$ ulimit -d unlimited (bash shell)
(As a bash user I was looking for a limit command there which I didn't
find). But I believe my big mistake was to believe that if I changed as root
a parameter like this that it would be systemwide, but it is only for
the process. Changing the data segment size to unlimited resulted that
Matlab then worked on those files which failed before.
Arno pointed out that I could change the parameters with sysconfig
or dxkerneltuner or should add at the end of /etc/sysconfigtab
proc:
max-proc-per-user=256
max-threads-per-user=512
... usw. Variablen stellen
Sorry for sending this simple problem. But I learned a lot...
Regards
Otto
PS: But...the problem is still not yet solved totally. The data segment
size is now "unlimited" i.e. equivalent to the main memory of the
system. And of course my user had a just large data file (280MB) and
again we got "out of memory"... I will provide a followup of the issue.
Original Problem was:
A user working on large data files (about 100.000 data points) gets
often the message from MATLAB "out of memory" and the hint for the
sysadmin to increase swap space. The system has 512 MB memory and
about 1,5 GB swap area(SW: Unix V4.0D, Patch 4). When the user was
running Matlab there was no one else on the system and swapon
showed available space 99%.
...
The only idea I have so far is a too small data seg size (s. system
parameters below). But this is obviously hardcoded to 131 MBytes.
May I change this with sysconfig and how large is reasonable? Or
where else can I increase it?
-------------------------------------------------------------------------
| Dr. Otto Titze, Kernphysik TU, Schlossgartenstr. 9, D-64289 Darmstadt |
| titze_at_ikp.tu-darmstadt.de Tel: +49(6151)16-2916,FAX:16-4321|
-------------------------------------------------------------------------
Received on Wed Aug 11 1999 - 16:56:40 NZST