Found it : compiled my test-program with "gcc", so started searching if
"gcc" needs to produce code specially for 64-bit environments -"-m64"
option

gcc -m64 -o memalloc memalloc.c

Now it runs as a glitch over the 4GB 'boundary'

Rob

Rob De Langhe wrote:

Hi,

recently re-installed E25K domains with Solaris-10, and my users
complain that their application fails to load a big chunk of data into
memory to produce later an output file with reworked data. When they
ran their program on the same box but under Solaris-8, such output
file was about 250MB in size, so not spectacular.
While I monitor the size of such process (in Solaris-10) during its
loading of data, it exits just around 4GB of virtual memory size, with
error "out of memory".

So I reproduce this, with a small C-program that calls "malloc()" for
chunks of 200MB each. It does so till exactly 4GB, than terminates
with "Not enough space"

My server has 128GB of RAM installed, 4GB of swap.
While I run my tiny (but growing) program, the swap remains entirely
unused ("swap -l" reports all 4GB free), and RAM decreases from
around 82GB free to 78GB free, then back (when the space from my
program is returned).

I searched all around the kernel params, docs.sun.com, and Google, but
nowhere anything that resembles a 32-bit behaviour to limit memory
size per process to 4GB.
Yes, my machine runs in 64-bit mode ("isainfo -v" reports 64-bit
sparcv9 kernel)
My user (as any other user) has % ulimit -a
virtual memory (kbytes) unlimited

Any hints or tips?

Rob
--