I gave this presentation during the seminar of the "Large Scale System" department of IRISA. It presents some classical methodological pitfalls that PhD students (and others) will face when experimenting with such systems. It speaks of SimGrid, but it's not centered on simulation at all.