FEAP User Forum

FEAP => Parallel FEAP => Topic started by: sicheng6699 on May 29, 2017, 08:00:50 AM

Title: comparison of run time
Post by: sicheng6699 on May 29, 2017, 08:00:50 AM
Dear Prof. Taylor, dear Admin,

to test whether parFEAP could improve (accelerate) out simulation, a simple test has been made, which is a implicit dynamic analysis of a cantilever beam under self weight.
In the attached files you can read from the Ofiles, that the sequential run used 53 sec., and the parallel run used 63 sec. (with np=4).

First question form my side, should the optimal speed up be 4? Or precisely speaking close to 4?

If true, there are some issues that might lead to the problem, but I am not totally sure and it is also very difficult for me to test all. The general informations that I can give are as follows:
- feap 8.4
- petsc-3.4.3
- intel fortran compiler
- run with command "mpirun -np 4 ./feap  Itest_par_rev4_0001"
- implicit ?
- dynamic ?
- utan ?
- problem still to "small" to be parallelized?
- (our Institute Makefile, it could be provided if needed)

Thank you for any help or suggestion!
Title: Re: comparison of run time
Post by: sicheng6699 on May 29, 2017, 08:02:47 AM
sorry, this has to be separately uploaded...
Title: Re: comparison of run time
Post by: Prof. S. Govindjee on May 29, 2017, 11:59:11 AM
Your problem is trivially small to see any advantage.  There are only ~200 equations per processor.  This needs to increase a at least 2 orders of magnitude before you see any advantage.  Try something with maybe 200k total equations or even 800k.
Title: Re: comparison of run time
Post by: sicheng6699 on May 30, 2017, 12:00:07 AM
I understand what you mean, but for our research, we are not dealing with "huge" models that take days for computation. Maybe I should try something else to accelerate.
Thanks a lot for your reply!