New to MPI, currently I dont have a super computer (yet) to play with, one thing I dont know is, how is DLL/SO handled in cluster computing? Can we use DLL/SO the same way we used them on a PC?
Why there are so few supercomputer that support Windows HPC OS instead of Linux? Currently I think programming software on windows is much easier thanks to visual studio, but too bad it looks like for cluster-computing, linux is the first choice at the moment.
To add to what you've already been told in the comments to your question:
a) I think that one part of the reason for Linux's dominance in the world of supercomputing is that it is open-source. Once you've developed the hardware for your 10^6-CPU behemoth you might want to tweak the o/s a bit to optimise it for the particularly cool features you've designed and built in.
b) If you like developing under Visual Studio but want to run your codes on a supercomputer you can do what I do: develop code under Visual Studio and run code on a supercomputer. (Actually, my employers don't really have a supercomputer, just a very large cluster or two but meh.)
I have a desktop PC with twin quad-core CPUs, an installation of VS, the Intel Fortran compiler and a variety of Intel libraries such as the MKL. I also have an installation of the Intel MPI development kit and run-time for the PC so I can run MPI programs on my PC while I develop them.
Then, when I want to move the codes into production, I make them on the cluster (where we have the same compiler and libraries) and away they go. We stick rigidly to language standards, so we don't build in portability problems.
So, I don't see why you should have to ditch Visual Studio if you want to work on supercomputer codes. Sure, you'll need to acquire some familiarity with make and a programmer's editor on the Linux side of things, but they're no more difficult to come to terms with than their equivalents once were for you on Windows.