On Mar 24, 2014, at 2:54 PM, Qiyue Lu <[email protected]> wrote:

> Dear All:
> 
> I am solving a series of matrix which are from the same FEA model but have 
> different number of meshes. In petsc, I use CR as the solver and 
> hypre-boomeramg as the preconditioner. It works for a small system with 2 
> million DOFs. Also works for a system with 6 million DOFs. But always fail 
> with an error ' memory out of range' for a 10 or 20 million system. 
> 
> Additionally, with solver CR + preconditioner bjacobi, Petsc could solve 10 
> and 20 million and even large systems easily. 
> 
> Also, with using top command in linux by monitoring the memory usage, I saw 
> some peaks in the memory.
> 
> So, I am wondering, is this because the memory usage of boomeramg, that 
> larger systems always fail?

   BoomerAMG does require a good amount more memory than “simpler” solvers. 
There is nothing one can do about it except use more MPI processes with more 
total memory for larger problems. 

   Barry

> 
> Thanks
> 
> Qiyue Lu

Reply via email to