When cell number increase，the RAM continue to increase until all memory is occupied

MFiX How to
#1

Hi everyone, as the title say, when the mesh number is 2828112, the simulation runs well.But when i refine the mesh to 56 * 56 * 224, the RAM continue to increase until all memory is occupied, and then the program crashes.
attached my simulation data, it is a simpel case in cyc boundary with heat transfer. my totao RAM is 125G, CPU`s core is 56.
usr_sources.f (6.3 KB)
cyc_dropp.mfx (14.8 KB)

#2

Hi -

I ran this case for several hours and monitored the memory usage (resident set size, or RSS) with the ps command. I did not see any memory usage increase over time:

plot2

it’s consistently about 1.6GB. This is with ijkmax = 56,56,224

Can you tell me more about the system you ran this on? Did you compile the solver or are you using default? What operating system, version of Fortran, etc?

#3

Thanks for your kindness.
I run this case in Centos linux release 7.4.1708，with Mfix version-22.1.1. And i compile the solver using:
build_mfixsolver --batch --dmp -j 24 FC=“mpifort”
gfortran is 4.8.5

#4

Ok. I was only testing the serial solver (single CPU)

How are you launching the solver process? How many nodes are you running on? What are your settings of nodesi, nodesk, nodesk ?

#5

i launching the simulation using ：
mpirun -np 56 ./mfixsolver -f cyc_dropp.mfx NODESI=2 NODESJ=2 NODESK=14
,however i also try other nodes and settings,like NODESI=2 NODESJ=2 NODESK=8、NODESI=2 NODESJ=2 NODESK=6…they all face the same result.

#6

plot3

I monitored two processes, PID 6994 was PE0 and 6998 was one of the other worker jobs. Memory usage for PE0 stayed constant, while the other fluctuates a bit but did not trend upward in 30 minutes.

Do you see a memory leak for the 2x2x2 decomposition?

What version of openmpi and gfortran? I’ve got

mpirun (Open MPI) 4.1.4
GNU Fortran (Gentoo 12.2.0 p1) 12.2.0
#7

I test the 2x2x2 decomposition. RAM still increased，thought the velocity of increase get slower.
my softwares version are followed,
Open mpi 2.1.6；
Gfortran 4.8.5；

#8

That is a very old Gfortran version. Can you try with newer GFortran and openmpi?

#9

image
image929×476 30.9 KB

thanks for help， i upgrade the gcc and openmpi as followed,but the Mfix still choose a 4.8.5 version.Do I need to reinstall the Mfix？
image
image821×92 5.66 KB

image
image819×96 5.84 KB

image
image815×66 3.83 KB

#10

That’s odd. Try doing make clean or deleting the build directory.