Has any one successfully calculated the drag coefficient of a cylinder in turbulence flows? I have tried several grids and different turbulence model, but failed to get the right result, it always differs from the experimental more than 30%. (At Re 1e5, Cd from experiment is 1.17)

It's very tough comparing simulation results to experimental results. You have to ensure you have eliminated all possible sources of errors in your simulation before comparing the results.

Try using structured hexahedral mesh (O-grids for cylinders should be OK I guess)with Y+ around 1 and many layers of prisms in the boundary layer, and use k-omega SST turbulence model. This will reduce numerical diffusion, and at the same time capture the boundary phenomenon as close as possible. Start your solving with first order to ensure convergence, and restart with second order discretization schemes for more accurate results. You should get quite accurate Cd predictions.

Hope that helps!