Hi,

My libmesh progam, which uses SNES to solve a nonlinear equation, had different 
convergence behavior when running on a linux desktop and on a cluster machine.

All other running conditions were double checked to be identical: It was the 
same program 
running as a single thread on both machines. Both machines were using Libmesh 
0.9.2.2.

My program was solving a simple elastic model with good numerical/physical 
behavior. (The 
model consists of a 5*5*5 cubic mesh with a total of ~600 degress of freedom.) 
For the 
nonlinear solve, I was using the Newton method and line search (the default 
settings in 
Petsc SNES).

The Newton method took 5 iterations to converge on my desktop (Ubutun 12.04), 
whereas it 
took 114 iterations to converge on the cluster. Moreover, both machines 
generated 
identical results in the first 2 Newton iterations, then they diverged from the 
3rd 
iteration.

I am wondering if such difference is common for Libmesh/Petsc programs running 
on 
different machines. Does anyone have similar experience of running programs on 
different 
clusters?

Thanks,
Dafang
-- 
Dafang Wang, Ph.D.
Postdoctoral Fellow
Institute of Computational Medicine
Hackerman Hall, Room 218
Johns Hopkins University, Baltimore, 21218
http://lagniappe.icm.jhu.edu/~dwang/

------------------------------------------------------------------------------
Rapidly troubleshoot problems before they affect your business. Most IT 
organizations don't have a clear picture of how application performance 
affects their revenue. With AppDynamics, you get 100% visibility into your 
Java,.NET, & PHP application. Start your 15-day FREE TRIAL of AppDynamics Pro!
http://pubads.g.doubleclick.net/gampad/clk?id=84349831&iu=/4140/ostg.clktrk
_______________________________________________
Libmesh-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libmesh-users

Reply via email to