DLPOLY4 / node/domain decomposition


Trying to run examples (here test01) from DLPOLY4 on a supercomputer, i get this (see below).
How to solve this ?
I guess it is linked with parallelisation but do not see how to fix it.

Many thanks in advance.
G. Dezanneau


It seems you are running over 80 (445) parallel processes. The cutoff for your specified interactions is 12A. DL_POLY uses a domain decomposition parallel strategy: the system as a whole is broken into domains, each of which is treated by a particular parallel process. I’ll guess that your system is relatively small compared to the cutoff you have specified?

Link cells are allocated in each domain, and the number of these cells is determined by the interaction cutoff and the size of the domain. If you have too few link cells per domain (as in your case), the strategy is suboptimal - because you spend a lot of time sharing “halo” data (data on the boundaries of each domain) with other domains.

I’d recommend reading section 8 or the manual (specifically part 8.2.6 - “Simulation Efficiency and Performance”) for more detailed information about this.

Your problem is a natural consequence of using a domain decomposition strategy to try to speed up the simulation of a relatively small system: more processes isn’t always better. Generally, in this situation, the options for increasing parallel efficiency are:

  • use fewer parallel processes (domains larger)
  • decrease the cutoff (more link cells per given domain)
  • increase the system size (domains larger)
    Hope this helps, best wishes,


You should probably try a bigger system. Low efficiency means that you have too many cores for the system size. You may want to also check your cut off, there are some constrains to ensure correct integration of algorithms. I find that the explanations and instructions of the errors in the manual are very useful.



Eva Zarkadoula, PhD Physics (she/her/hers)

Materials Science and Technology Division

Oak Ridge National Laboratory, USA