Let the serial run time of an n-point FFT computation be tcn log n. Consider its implementation on an architecture on which the parallel run time is (tcn log n)/p + (twn log p)/p. Assume that tc = 1and tw = 0.2.
1. Write expressions for the speedup and efficiency.
2. What is the isoefficiency function if an efficiency of 0.6 is desired?
3. How will the isoefficiency function change (if at all) if an efficiency of 0.4 is desired? 4. Repeat parts 1 and 2 for the case in which tw = 1 and everything else is the same.