Sample size for SPC

For my understanding, few SPC training covers sample size selection. For example, how many data should there be in a subgroup for, say Xbar-R chart? Usually the answer is: 3-8 never>10. But why? any mathematic background for the 3-8?
ACtually, i know that the sample size ought to be different for different process control. If one want to apply SPC effectively,he should determine the sample size correctly, based on the true process performance,the alfa/belta risk. But HOW TO? I'm asking help. Appreciate if you can show me a step-by-step method or direct me to some relative article or website. Thx