Dear Users and developers,
Recently, I am facing a strange problem with abinit. I have installed abinit 9.8.4 on ubuntu22.04 under VMware workstation using 12 processors.
I can reproduce many tutorials so far I checked using “mpirun -n 12 abinit inputfile.abi > log” and it works fine and faster. I just copied the input file and pseudopotentials from abinit test folder and ran it.
However, when I write my own input file (myfile.abi) and run it, it stays forever at 1st iteration. I don’t know what is wrong! It is very small calculation should take only few minutes.
BTW, I have run my own input file (myfile.abi) in another machine and it works fine. Therefore, it is difficult for me to find the problem. At least it is not a problem with input file.
Could you please check the issue? Here I attach the tutorial file and my own files.
tbase3.abi (2.0 KB)
tbase3.abo (28.1 KB)
log_tbase3.log (32.8 KB)
log_myfile.log (23.0 KB)
myfile.abi (830 Bytes)
myfile.abo (14.0 KB)
Thanks,
Rajesh
Dear All,
I think the problem was with number of processors assigned for VM workstation. When it is almost same number of guest OS, it works very slow. When I reduce the #procrssors then it works. BTW, does it matter if I compile abinit with “make -j8” and run the code with “mpirun -n 10 abinit xx.abi” while the number of processor is 12?
Regards,
Rajesh
Dear Rajesh,
I suggest you follow the tutorial on parallelism. I should provide guidance.
As for compiling, this is a separate thing as running the code. You can compile with any number of processors without affecting the final executable. Now the number of processor you give in the mpirun -n part of running the code is also independent of the number of total processors of your machine, as long as it is smaller. This is the number of CPUs that will be used for your calculation.
Hoe this helps,
Best,
Olivier
2 Likes
Dear Olivier,
Thanks for the clarifications. For the 1st problem, it seems to be related with processors sharing between guest and host. I have one socket and 16 cores in windows 11 enterprise. When I give 16 processors to my guest VMware workstations and I run mprirun -n 16 or 12 it takes long time to compute a 2 minutes job. But when I assign 8 processors to my guest VMware and run mpirun -n 8 it takes 2 minutes. Therefore, I guess when VMware takes all the 16 processors virtually nothing left for host windows and it is difficult for her to run the VMware itself.
I had to find a balance where both guest and host works fine but much faster way which is 8 processors to VMware.
Best regards,
Rajesh