[gmx-users] Segmentation Violation

[James O'Dell]

I have not done detailed scaling tests yet but I have tried on
2,4,6,8... and it seems to get segmentation violations on anything
greater that 8 processors.

A stack trace from the debugger seems to indicate that MPI signaled
the error. Could it be that there is simply so much communication
traffic that MPI's internal buffers simply overflowed?

Jim


On Wed, 2003-03-12 at 12:34, Anton Feenstra wrote:
> James O'Dell wrote:
> > I have a simulation that I can run on a single processor or on an 16
> > processor cluster. The simulation seems to run just fine on the single
> > processor. ON the 16 processor cluster, it gets a segmentation fault
> > almost immediately after loading. I'd appreciate any pointers on how to
> > debug this.
> 
> Try it stepwise, i.e. first try a single processor run on your cluster,
> then 2, 4, 8. (While you're at it, you can also keep track of wallclock
> times used so you get an idea of the scaling.)
>