run getdp backstage

[Christophe Geuzaine]

Lin Ji wrote:
> 
>    Also, can you give me an idea how much is the problem size increased when
> I change the interpolation oreder from 1 to 2 on the same meshing? 

It depends on the following factors:

1) matrix
- the bandwith B of the matrix increases when you go from order 1 to
order 2 (the actual number depends on your particular mesh: just
consider how many elements touch one node + how many touch an edge)
- the number of equations N increases from #nodes to #nodes+#edges (that
for your case: see the FunctionSpace definition)
- all in all, this permits to compute (approximatively) the number of
nonzero  elements in the matrix:
D1 = N*B.

2) preconditioner
- depends on the method. Check the explanations in SOLVER.PAR. With
ILUT/P, you store D2 = (2*NbFill+1)*N.

2) solution vectors:
- if you don't do the post-pro at the same time as the solving, 3
vectors are stored during all the processing -> D3 = 3*N.

3) additional GetDP structures (of course...): the mesh, the dof
structures, etc... For big cases, you may neglect this (mostly integer
stuff -> maybe D4 = 3*N?).

Total memory = (D1+D2+D3)*sizeof(double) + D4*sizeof(int)

> If I half
> my meshing size, how much more memory does 'getdp' need for the 1st and 2nd
> order respectively?

See above...

-- 
Christophe Geuzaine

Tel: 32 (0) 4 366 37 10    http://geuz.org
Fax: 32 (0) 4 366 29 10    mailto:Christophe.Geuzaine at ulg.ac.be