The meshing method can differ for each geometry to get reliable results. Usually, higher mesh element concentration is used near sharp corners or relatively smaller geometrical regions (see figure). Quadratic order meshing might be more accurate than linear order meshing. Checking Orthogonal quality of elements (higher the better) and skewness of elements is necessary before proceeding for solution iterations.
I think there is no a priori "perfect" mesh for a simulation task. There will be different meshs that solve the problem with the same quality, still respecting Yatish's advice. You will have to compare different ones until you get confidence in the model and the mesh. This is depending on experience. A convergence study might help - stepwise decreasing the element size (h-convergence) or increasing the element order (p-convergence).
For precision calculations a fine mesh is required, that is very small and for calculations that do not require precision equivalent to a thicker mesh is sufficient, keep in mind that the finer the mesh is, the more calculations the computer will perform and therefore more resources will conserve her software.