question about Loop time in LIGGGHTS

there is an easter egg in liggghts 3.1.0, insert at the beginning of your input deck

modify_timing on

or

modify_timing verbose

to get a detailed output which fix consumes most of the time. For me it's often move mesh because some expensive double operations are needed to move all edges of the mesh-trianges in space.
You can speed it up with a better mesh (only needed details, as coarse as possible, equilateral triangles to allow perfect parrallisation ...).

For the standard move mesh example the output looks like this:

Loop time of 16.25 on 4 procs for 25000 steps with 1257 atoms
Pair time (%) = 5.22148 (32.1323)
Neigh time (%) = 0.0340779 (0.209711)
Comm time (%) = 0.156253 (0.961562)
Outpt time (%) = 1.11314 (6.85014)
Modfy time (%) = 9.53624 (58.6847)
Other time (%) = 0.188761 (1.16161)
Fix m1 property/global time (%) = 2.85072e-007 (1.75429e-006)
Fix m2 property/global time (%) = 0 (0)
Fix m3 property/global time (%) = 0 (0)
Fix m4 property/global time (%) = 0 (0)
Fix m5 property/global time (%) = 0 (0)
Fix 1 nve/sphere time (%) = 0.191115 (1.17609)
Fix 2 gravity time (%) = 0.0480044 (0.295412)
Fix boxwalls_x1 wall/gran time (%) = 0.0613983 (0.377837)
Fix history_boxwalls_x1 property/atom time (%) = 2.05266e-005 (0.000126318)
Fix boxwalls_x2 wall/gran time (%) = 0.0543399 (0.3344)
Fix history_boxwalls_x2 property/atom time (%) = 3.04099e-006 (1.87138e-005)
Fix boxwalls_y1 wall/gran time (%) = 0.0946886 (0.582701)
Fix history_boxwalls_y1 property/atom time (%) = 5.03665e-006 (3.09949e-005)
Fix boxwalls_y2 wall/gran time (%) = 0.0894974 (0.550755)
Fix history_boxwalls_y2 property/atom time (%) = 3.61109e-006 (2.22221e-005)
Fix boxwalls_z1 wall/gran time (%) = 0.420776 (2.5894)
Fix history_boxwalls_z1 property/atom time (%) = 2.66079e-006 (1.63741e-005)
Fix boxwalls_z2 wall/gran time (%) = 0.00205816 (0.0126656)
Fix history_boxwalls_z2 property/atom time (%) = 2.56585e-006 (1.57899e-005)
Fix cad1 mesh/surface time (%) = 0.250621 (1.54229)
Fix contacthistory contacthistory time (%) = 0.0040767 (0.0250874)
Fix bucket_wall wall/gran time (%) = 0.284114 (1.7484)
Fix wall_neighlist_cad1 neighlist/mesh time (%) = 0.0050307 (0.0309583)
Fix n_neighs_mesh_cad1 property/atom time (%) = 1.13085e-005 (6.95909e-005)
Fix tracker_cad1 contacthistory/mesh time (%) = 0.00342052 (0.0210494)
Fix movecad2 move/mesh time (%) = 8.02705 (49.3974)

you can see that movecad2 is expensive.

cheers,
Christian.

Hi Zyan and Christian,

yes modify_timing was introduced as an "experimental feature" some time ago by Richard Berger. But I think now it should be stable enough, I'll add it to the documentation for the next release

Cheers
Christoph

I was under impression that "pair time" includes the time take for calculation of overlap and then calculation of forces.
However I dont see it happening that way. The time reported at the end is much lower for "pair". All the actual time is added to "other".

Anyone else observes similar behavior?

Thanks