Cannot increase stacksize beyond stacksize(33554400)
Dominik Kehl
for some reason I can't increase the stacksize to more than
stacksize(33554400), corresponding to about 268 mb. That limit does
not change, no matter how much I increase the size of the swap file.
I even just got an additional 128 mb of RAM, so I have at least 256 of
RAM available now, but still cannot increase the stacksize beyond that
limit. Is that a general limitation of Scilab's stacksiz? Or does it
just occur on specific systems? I am running Scilab under Win95 and
Win85.
I am in desperate need for a greater stacksize since I want to compute
some large vectors, which I cannot now, unfortunately.
So, any help and information is greatly appreciated.
Dominik
Lydia van Dijk
On Mon, 13 Aug 2001 17:44:49 GMT, Dominik Kehl
<ke...@rz.uni-potsdam.de> wrote:
> for some reason I can't increase the stacksize
> to more than stacksize(33554400),
> corresponding to about 268 mb. That limit does
> not change, no matter how much I increase the
> size of the swap file.
Looks like a wanna-be OS to me. :->
On this ia32 Linux-2.4 box, stacksize() can
allocate memory until it runs into one of the
ulimits,
data seg size (kbytes) 256000
max memory size (kbytes) 256000
virtual memory (kbytes) 512000
which results in a maximum arg of 56000000
(without ulimits it would hit the 3GB limit
imposed by the kernel (I guess, but have not
tried.)) Scilab then has the following memory
footprint:
$ ps vw
PID TRS DRS RSS %MEM COMMAND
1343 404 1867 1368 0.1 -bash
1353 959 7368 6368 0.8 emacs
2591 69 2154 1268 0.1 rxvt -e scilab -nw
2592 404 1707 1072 0.1 /bin/sh /site/X11R6/bin/scilab -nw
2599 13 1054 500 0.0 /site/X11R6/src/scilab-2.6/bin/geci ...
2600 3874 443949 2092 0.2 /site/X11R6/src/scilab-2.6/bin/scilex ...
2629 76 2503 1000 0.1 ps vw
> Is that a general limitation of Scilab's
> stacksize?
stacksize() calls scimem in
SCI/routines/default/scimem.c which in turn
relies on malloc(3). Thus, you are at the mercy
of your MM system.
> I am in desperate need for a greater stacksize
> since I want to compute some large vectors,
> which I cannot now, unfortunately. So, any
> help and information is greatly appreciated.
Any 32bit system will surrender at 4GB total
process size -- usually before that limit.
Switching to a 64bit arch (AXP, IRIX?) to me
seems like the fastest way to get rid of your
problems.
-lvd
--
Lydia van Dijk
Hammersmith Consulting, web: www.hammersmith-consulting.com
Remove all 'NOSPAM.' from Reply-To to get correct reply address.
Dominik Kehl
Dominik