Use of GB in parallel and/or with all-vs-all kernels needs a mention in the manual
Currently, I think the manual makes no mention of any of the details of running Generalized Born calculations in parallel, nor about the use of the all-vs-all loops.
In particular, GB simulations without cut-offs using those loops silently trigger PD, which is confusing when PD-related error messages subsequently result. There's been at least one gmx-users thread because of it.
I think the correct person to whom to assign this is Per Larsson, but I don't think he's currently on the Redmine manual project. I'll email him so it comes to his attention.
#4 Updated by Rossen Apostolov over 8 years ago
- Status changed from New to In Progress
- Assignee deleted (
Below is what Per sent me about it. Where should we put that though, in the mdp-parameters section?
"Note: In some cases it is preferable to run implicit solvent simulations with longer cutoffs than for explicit solvent. Gromacs therefore comes with the option to run with infinite cutoffs using special optimized all-vs-all loops for implicit solvent. These loops are silently triggered by setting all relevant cutoffs to 0 in the mdp-file
nstlist = 0
rlist = 0
rcoulomb = 0
rvdw = 0
rgbradii = 0
It is possible to use the all-vs-all loops for parallel computations, but that will trigger particle decomposition rather than domain decomposition.
Parallell all-vs-all calculations are quite fast for small to medium sized systems, but performance will be drastically reduced as the system size is increased.
To run a parallel implicit-solvent calculation with domain decomposition, the cut-offs have to be finite."
#5 Updated by Justin Lemkul over 8 years ago
It seems to me that this discussion is relevant in section 3.16, where PD is introduced. It is stated in section 3.15 that particle decomposition is useful "for a few special cases." I think that's what's being referred to. It would also be worth mentioning the fact that PD and all-vs-all kernels are not just relevant to GB simulations, but also those done in vacuo. Perhaps the text can be generalized in that way, with forward references to sections regarding GB simulations and the appropriate .mdp settings.