i agree with you, i am also observing that this has considerably slowed down my code, i am thinking in direction of improving efficiency.

Are you using some public library? I'm trying to speed up my code, and I started using MTL and ITL for matrix storage and linear solvers. They're both written in C++ and, if you know the language, you don't take too much time to learn how to use it. The addresses are http://www.osl.iu.edu/research/itl/ and http://www.osl.iu.edu/research/mtl/

no i am not using any liberary, i had already looked at mtl, perhaps i downloaded the source too, but i liked DEAL better, if you are working with finite element , have a look at this, you have to google for that, for me , i have three choices , once is the solver based on graph structure with edges-nodes (areas and cvs), another is matrix structure embedded and lastly one more sparse matrix which is supposed to be fast. actually i am trying to get things right first , then i will work on making is fast, so there are lot of areas where i could have made code faster, but i left that delibrately. further these lib does not make sense for me much because the sole purpose, is to write multigrid based solver (these mostly provide, linear solvers like BiCGStab etc, Deal provides multigrid though), futher i wish to wirte meshless type of solver of matrix for that will be little different than usual, so for me lib still do not make much sense.

I've downloaded deal some time ago, but it works with adaptive mesh refinement, and at the moment I'm working with unstructured grids without any remeshing. That's the only reason why I didn't go on with it. I'm working with control-volume finite element methods for incompressible flows (step-wise constant weight functions), which some call unstructured finite volume, but I'm doing it in a finite element fashion (element matrices -> global matrix, and so on). I think AMR is a good TODO in my future research, and I'm thinking of using deal to work with immersed finite element method in the future.

No,i didn't heart anything about it since this time, what is it?

In case you are not able to use a search engine:

http://www.cfd-online.com/Forum/news.cgi?read=1845

well one thing at last, that my unstructured grid solver also has started to run, now i can peacefully think about implementing multigrids. the probelm was tracked down to a very silly but subtle mistake, the wt was calculated as dist(P to face center) / (sum (dist P to face center plus Neighbour E to face center), this term is okey when grids are nearly orthogonal (so it was running well with structured meshes), but this is not good approximation when the meshes are non orthogonal. once this is fixed things are fine.