Hi everybody,
thanks to all who responded
Donald Bovee <bovee_at_amath.washington.edu>
Hamblet, Robin" <rhamblet_at_water.ci.portland.or.us>
Erik Piip <piip_at_merl.com>
Lisa Kim <kim_at_zso.dec.com>
Michael Gasperi <michael.gasperi_at_uni-klu.ac.at>
Mayfield, Rob" <Rob.Mayfield_at_compaq.com>
Concerning my last question from the different hints and
experiences it looks now to me
- data segment size
Obviously this can be increased over the main memory size
up to memory + swap space (in some cases even above the physical
size?)
(I was somewhat confused because "ulimit -d unlimited" set it
to the main memory size. Therefore - as an primary VMS user -
this was suspect to me (like many things in Unix).
- even the stack space can obviously go over the whole memory
(Can't imagine what consequences this may have in Unix if
everything is unlimited. This is much clearer for me in VMS.)
- simplest way to change this sizes permanently is obviously if
one is adding those at the end of /etc/sysconfigtab
proc:
per-proc-data-size=1073741824
etc.
and reboot
- Changing at run-time
(/sbin/sysconfig -r proc per-proc-data-size=1073741824)
is obviously not possible. (not dynamic parameter)
Now the situation is improved. But I still find some interesting things
which I don't understand and will put in a another followup.
Thanks and regards
Otto
Original Question
>
>by increasing the data segment size to unlimited I get 1048576 kbytes
>now as limit (approx. the size of the memory of my system(512MB).
>Which is not enough in certain cases.
>
>Because the swap space is 1,5GB - is it reasonable to increase this
>size further on or can these areas not be swapped?
>If it can be swapped - there could another problem arise. The application
>(Matlab) needs contiguous memory. If it calls for contiguous memory
>could this be considered as such if it needs swap space?
>
Some interesting experiences with Matlab:
Donald Bovee <bovee_at_amath.washington.edu>
Matlab stores variables on the stack, so try increasing the stacksize
parameter. Matlab never deletes anything during a session so if your
researcher uses one large data file after another assigning a different
array name to each, the effect is cumulative. Use the 'clear' command to
get rid of old arrays and make space available.
Erik Piip <piip_at_merl.com>
We have an 8400 (8 440mhz CPUs, 2 Gig memory) that is being tortured by
matlab users.
We ended up added two 10,000RPM 9 gig disks as swap. This plus the changes you
had helped a lot!!!!
>From the top enclosed, you can see they have used 3 Gig of the swap
right now. I've seen it up to about 9 Gig of swap used..... These
matlab uses always want more ;-) Time for 8X EV6 CPUs!!
(The top showed 6 matlab users with 99% cpu usage)
Received on Fri Aug 13 1999 - 07:38:56 NZST