next up previous contents
Next: Performance tuning MPI 1 Up: Performance Analysis Previous: Scalability of Parallel Architectures   Contents

Assignment:

Solve 2 questions.
  1. Starting from the equation for the speedup factor given by

    $\displaystyle S(n)=\frac{1}{f+\frac{1-f}{n}}
$

    show the inequality that relates the fraction of serial computation, f, and the number of processors employed, $ n$, if a 50% efficiency is to be achieved.
  2. Consider a parallel architecture built using processors each capable of sustaining 0.5 megaflop. Consider a supercomputer capable of sustaining 100 megaflops. What is the condition (in terms of $ f$ ) under which the parallel architecture can exceed the performance of the supercomputer?
  3. Consider an algorithm in which ($ 1/ \alpha $) th of the time is spent executing computations that must be done in a serial fashion. What is the maximum speedup achievable by a parallel form of the algorithm?


Cem Ozdogan 2006-12-27