This paper presents the structure of program execution traces. This study extends previous works from many researchers aiming to quantify ILP. Our goal is to understand the general structure of a run and the instruction parallelism it offers. This structure is mainly made of two opposing parts. One has a high ILP which can infinitely increase when the trace lengthens and the other has a very low ILP (close to 1) which can extend infinitely with the trace. The first part comes from loops controls (loops with a counter set from a constant). The second part comes from functions parameters transmissions. Our results show that, without any modification of the ILP distribution, a processor must conciliate a high resource demand at the start of the run and a nearly sequential execution at the end.
