I am simulating a leaching process by the CFD-CGP method. There are only 17904 particles and 5600 cells. Chemical reaction happens on every particle. I am using 30 cores and it takes 48 hours to simulate 0.6 s. This speed is even slower than running on my desktop computer by a single core.
I checked the log file and the dt is normal, at around 0.001.
leach.zip (13.4 MB)
The attachment is the case with the results.
When I use multiple core the speed is even lower than single core.
I think we would need more information about your system to answer this question. What are the specifications of your parallel machine - what is the Linux version? DMP or SMP? Are you using Virtual Box, Linux workstation, or a supercomputing cluster at your university?
Hi, I am using a supercomputing cluster supported by government. Here is the introduction. Gadi User Guide - NCI Help - Opus - NCI Confluence
Moreover, I found if I run it from the terminal the case runs quicker and the time for dem and fluid solver came to normal. But when I submit it to the cluster, the speed slows.
Interesting. I’ve not had this issue on a supercomputing cluster before, so hopefully someone else can provide more specific guidance, but a few things you could try:
- Have you run any of the tutorial cases both in serial and in parallel on the supercomputer? If you have the same problem (simulation speed is slower in parallel), this could be an indication that something is not working as expected with your openMPI installation or its interface with MFIX.
- Have you experimented with the number of cores you assign to the simulation? For example, try just 15 or 5 and see if you are still running slower than the serial simulation.
Most supercomputers also have a support staff that can help look into how your program is using resources like memory, if you haven’t tried that already. They may be able to identify a bottleneck in resource usage that would explain the slower parallel run.