You need a conversion factor PU <=> LU (PU = physical units, LU = lattice units).
For example: 1 lattice spacing shall be 200 micrometers. This sets your length scale and you know that the cylinder diameter of 20 mm would be 100 lattice nodes.
This conversion factor has a priori nothing to do with your Reynolds number, but be careful:
1st: A larger Reynolds number usually requires a larger resolution, i. e. more lattice nodes.
2nd: The lattice constant (length of a lattice cell) should not be too small, because the simulations would take ages.
So, my advise: Estimate how may lattice nodes you would like to have in total (e. g. 2 million) and then calculate how many that makes per axis (e. g. 100 x 100 x 200, of course different in your case).