How to systematically figure out `how much of memory overhead is being used in spark job`?

Question

Is there any way on how much memory overhead a spark job is using at a point in time? I know we can come to a number by hit and trial but would like to go systematically for som