Hi,
Sorry, but the kind of memory management that you are talking about is not
existant in lam :-). We clean up all the memory which we allocate
internally in our libraries. But we are not responsible for the memory
that was allocated by the user :-). Segfault behaviour is completely
dependant on the OS (which in most cases frees up the memory). We do not
do anything special in those cases. LAM daemons keep running when the
process goes away.
Hope this helps,
Anju.
> Hi all!
> I am in charge of a scientific code which runs on about ten
> work-stations using lam-mpi. The os on all nodes is a Linux RedHat9
> (2.4.xxx). LAM/MPI is 7.0.2
> This code has *enormous* memory leaks (but I didn't write them!!!):
> nothing (or very little) have been done to clear the mallocated slots
> and Valgrind yelllllls each time it is run on this program.
> Are the LAM deamons capable of dealing with memory leaks on remote hosts
> : are they in charge of giving back the memory they got for running
> their process to the remote os (Linux in my case)? If yes, it seems to
> me that each time a lam process finishes, the os will wipe out the used
> memory and make it available again. Or perhaps is it when I lamhalt from
> my master node? What happens to my deamons anyway when my program
> segfault on a remote host. Does LAM handles this too?
>
> Thanks for your answers. If there is already a faq or doc I can read to
> get these answers, do not hesitate to tell it, will you?
> Thomas
>
> _______________________________________________
> This list is archived at http://www.lam-mpi.org/MailArchives/lam/
>
|
