Hi all - released today an RCB (recursive coordinate bisectioning) option
for the static and dynamic load-balancers in the code. If your
simulation could benefit from better load-balancing, you
may want to try it out:
doc pages and pics/movie:
There is also a new comm_style tiled to support it:
Basically RCB is a fast, parallel way to assign equal
numbers of particles/processor by creating an irregular
tiling of the simulation domain, where each processor
still owns a sub-box that is a 3d brick. This makes it
easy for most other calculations in LAMMPS to work
Watching the toy movie link above will explain the idea
better than I just did.
No support yet for triclinic boxes or PPPM/MSM long-range
solvers, but that’s on the todo list.