"Is there a way to reduce computational time without compromising in accuracy ?" > would be a million dollar idea
i would recommend you start with a much coarser mesh, see what you results look like
then bump it up a bit see how much they change
if htey don't change significantly you may not need 3 million cells to solve the problem
35hrs for a 3 million cell problem seems like a lot, what are teh computer specs?
135hrs for 3 million cells seems very wrong, what are you trying to use as your convergence criteria?
Computer specs are pretty good. Its Intel Xeon 3.7 GHz with 8 core and 32 GM of RAM. Convergence criteria was set automatic. Geometry check was fine but Could it be a problem with my solid modeling? (too many solid structure inside with some of them imported from CAD) Picture attached
from a model perspective, only suggestion would be to simplify the people further
otherwise the model looks ok
remember that you don't have to wait for full convergence, you can stop it early
average parameters might be challenging to converge fully with a turbulent environment. maxes are sometimes better. (Even if average is your goal)
take a look at the actual plots to see how much it is actually changing