Hi,
While running simulations on HPC with multiple nodes, simulations are taking around 12hrs for 2crore time steps loop. Tried with 1024 cores , 640 cores etc.,
Each node has 16 cores and 64 gb ram.
Is there any command needs to be added in script file or any module missing in PBS file?
Please have a look at the attached files.
| Attachment | Size |
|---|---|
 LIGGGHTS script file | 5.19 KB |
| PBS file | 390 bytes |
paul | Sat, 12/30/2017 - 10:53
> #processors 4 4 2
> #processors 4 4 2
Uncomment this line and adapt it to the numbers of cores used / geometry.
It has been taking such a long time b/c you haven't enabled parallelization.
I do not know whether your PBS script is correct, but I am concerned b/c there is no machine file being created or passed to mpirun.
- Paul
shivagoud | Wed, 01/17/2018 - 05:29
Static domain is the issue
My simulation box is occupying 860 litres whereas my blender is 50 litres of which particles occupy 25Litres only. I think in LIGGGHTS-PUBLIC the processors gird is static.
medvedeg | Wed, 01/17/2018 - 10:29
Hi Shivagoud,
Hi Shivagoud,
you are right, processor grids are static in LIGGGHTS-PUBLIC. Dynamic load balancing is a feature of LIGGGHTS-PREMIUM and not available in LIGGGHTS-PUBLIC.