Processing math: 100%

 

 

 

Upper bounds on speedup

Assume that almost all parts of a code are perfectly parallelizable (fraction f). The remainder, fraction (1f) 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?