Assume that almost all parts of a code are perfectly parallelizable (fraction $f$). The remainder, fraction $(1-f)$ cannot be parallelized at all.

That is, there is work that takes time $W$ on one process; a fraction $f$ of that work will take time $Wf/p$ on $p$ processors.

• What is the maximum possible speedup as a function of $f$?