Re: [ESPResSo] Self-avoiding random walk SAW

[Axel Arnold]

On Friday 13 August 2010 22:07:04 Mikheil Azatov wrote:
> Hi,
>
> I think I actually figured out why I had a segmentation fault when I was
> running it in parallel. It seems like "part->bl.e" does not exist if the
> particles is a ghost particle so I was getting a segmentation fault when I
> was trying to access part->bl.e[0], ... So I just added a check on particle
> being a ghost before doing that part and it seems to work fine now. The
> program runs without errors now but there's still a problem.  It stops in
> the middle of the simulation as if there is an infinite loop somewhere. So
> basically at some integration step it would just not go forward although
> both CPUs seem to work on 100%. I wonder if anyone had this type of problem
> or it's just something in my code.

Hi!

Chances are rather good that the problem is somehow related to your code, 
although in parallel programming, errors do not necessarily show up where they 
actually happen - corrupt data can mess up the communication. A deadlock in 
MPI would for example cause both CPUs to go to 100% and not doing anything 
anymore. To figure out, where your code hangs, you can either try the 
debugger, or you can define quite a number of debug output statements in 
myconfig.h, which then should show you where the processes hang. Try with:

#define COMM_DEBUG
#define INTEG_DEBUG
#define CELL_DEBUG

Probably you will have to redirect output, it will be quite a lot... All 
output is prefixed by the node number, but can be mixed up; lines from 
different CPUs next to each other don't necessarily mean that the things 
happened at the same time, but at least, you can see what both CPUs did last 
before hanging.

Many regards,
Axel

-- 
JP Dr. Axel Arnold Tel: +49 711 685 67609
ICP, Universität Stuttgart      Email: address@hidden
Pfaffenwaldring 27
70569 Stuttgart, Germany