I have a similar problem as Jacob (http://fias.uni-frankfurt.de/pipermail/espresso/2009-November/000483.html) with my system:
I have a system with 47,800 beads in a simulation box (85 85 85), where approximately 2000 are charged beads interacting with coulomb potential (P3M method). Also, I have harmonic, angle and Tabulated Week-Chandler-Andersen interactions.
One of the molecules is a very large chain of beads (> 2000), with
charged and no charged beads, where they are conected by harmonic
bonds.
I ran my first simulations with skin = 0.4 and different number of processors [4,8,16], however, due to the size of the system and the charges the simulation was very slow. So, I tried to increase
the number of processors to 32,64 but always I got the same message
(Bond broken....) and the system crashes even during the thermalization
(NVT + capped forces and No Electrostatics).
In my case the
maximal range of interaction comes from the electrostatic interactions,
and the cutoff radio is automatically optimized by ESPResSo using the
command: inter coulomb $bjerrum p3m tune accuracy $acc r_cut 0 mesh 0 cao 0
The only way I found to ran the simulations was increasing the skin to 3.0, but the simulations are still slow, almost as the case of nprocessors 16 and skin = 0.4.
Any advice or suggestion to improve the way I am currently running my simulations would be greatly appreciated. ============================================================================================================================== Besides, I have a problem when I want to restart a simulation from a checkpoint (NVT). The same situation is present with NPT simulations.
In the case of restarting the simulations, because we have a time-limit
in our cluster (the job is automatically killed) I need to restart the
simulations from the last chekpoint. But even when the simulation was running fine, when I tried to restart the simulation it crashes and complains about floating point exception.
The same problemm occurs for NPT simulations, but in this case the
problem is worst because for any system after the thermalizacion (NVT
and capped forces) the system crashes every time.