Roderick asked me some time ago to post the solution of the problem:
I just wanted to let you know, that we managed now to run CUDD on a Linux
maschine with more than 1 GB of memory. The main problem seem to be that CUDD
allocates its memory out of the brk region, which is about 900 MB. This is
from my point of view mainly a problem of glibc which could do better (also
the prereleases of glibc2.2 suffer from the same problem)!
The problem is manifested by the example programs I append (mem4.c is
allocating memory in smaller junks and glibc decides to do this within the brk
region, meaning that the virtual adress space available to malloc is limited
to about 1GB per process, than mallocc returns out of memory. The second
example allocates 2GB in one step, which is done by glibc using mmap() not
using the brk region (limit here is 4-1 GB = 3GB).
Since it was not obvious at the end for me, when memory comes from which
region, we decided at the end enlarge the brk region in the Linux kernel to 2
GB and reduce the mmap region to 2GB (thanks to Thomas) by
Changing
linux/include/asm-i386/processor.h from
#define TASK_UNMAPPED_BASE (TASK_SIZE / 3)
to
#define TASK_UNMAPPED_BASE (TASK_SIZE / 3 * 2)
This way we can use the complete 2 GB physical memory when running CUDD.
Rainer.
>
> We are currently evaluating Linux machines here and we have problems, when
> CUDD (the BDD library under VIS) is allocating more than about 1GB (in total).
> The malloc returns a NULL pointer although the machines are all equiped with 2
> GB RAM.
>
> Thus is anybody using VIS with the CUDD backend successfully under Linux with
> problems requiring significantly more than 1GB of RAM?
>
> We are working with Linux companies on the problem (so it is probably no
> trivial problem like wrong kernel configuration), but I am just curious if
> there is somewhere a working setup.
>
> Thank you.
>
> Rainer.
>
>
> --
> Rainer Dorsch
> Abt. Rechnerarchitektur e-mail:rainer.dorsch@informatik.uni-stuttgart.de
> Uni Stuttgart Tel.: +49-711-7816-215 / Fax: +49-711-7816-288
> Breitwiesenstr. 20-22 D-70565 Stuttgart
>
>
This archive was generated by hypermail 2b29 : Sun Nov 26 2000 - 09:35:51 MST