[nileshjrane]

Originally Posted by arjun

What you did is no use unless you can show that what you did could be used for everyone.
You made some assumption and seems to be working for your special case but it would not make me know how to use the power of thousands of processors for what i am doing.

You can still not invert matrix locally without communicating and you can not still invert matrix by ignoring few off diagonals and doing less communications. If it were true we would have developed lots of methods around it. What you are assuming is that you are the only smarty pants and all the others are mindless stupids. There is a reason we do things the way we do. And the reason is that people have found out that it is really not possible to just ignore few things here and there and make things work.

That was rude. Don't want to pollute the thread so I am backing off from this thread.

Just my last post here. I was talking about the algorithm which is developed by NASA and used extensively by them for their hypersonic flight designs, extra-terrestrial probes, reactive flows etc. So I am not considering myself "smarty pants" you see, neither did I say that others here are "mindless stupids". I am merely telling my observations/opinions. BTW something which is applicable for whole of supersonic and hypersonic regime is not that special case, now is it??

There are algorithms which are more "parallel friendly" than others. E.g. Krylov subspace solvers. The computational physics guys have been using them since years. They don't invert matrix at all. I have done some literature survey out of interest, and then derived the conclusion.

You can choose to ignore my opinions if you feel I am wrong. I did not enforced anyone to accept my views. I stand by my view, you stand by yours. But do it politely.