479 views

Let $f$ be the fraction of the computation (in terms of time) that is parallelizabl$e$. $P$ the number of processors in the system, and $s_p$ the speed up achievable in comparison with sequential  execution – then the $s_p$ can be calculated using the relation:

1. $\frac{1}{1-f-f/P}$
2. $\frac{P}{P-f(P+1)}$
3. $\frac{1}{1-f+f/P}$
4. $\frac{P}{P+f(P-1)}$

recategorized | 479 views

+1 vote

• Let execution time without parallelism is 1.
• If f is the fraction of parallel computation (1-f) time must be spend for sequential computation.
• f fraction of computation can be distributed among P processors; So it will take f/P time to complete computation.
• Total time for computation with parallelism = 1-f + f/P

speed up = time without parallelism / time with parallelism

= 1/ 1-f + f/P

// Correct me if I am wrong!

by Boss (32.5k points)
selected
0
@Shiva,what basis you've used for calculation?