Actually truncation error's components are local and propagated error, summation of these two gives the global truncation error as far as I know. Am ı wrong?
A smaller step size will result in a smaller error (difference between computed solution and exact solution), but will require more computational time. A larger time step has a larger error, but runs faster. So, the optimal step size is the one that gives you a small enough error for the amount of time you have available.