A parallel program is written in an MIMD environment so that the program can be
ID: 3855280 • Letter: A
Question
A parallel program is written in an MIMD environment so that the program can be broken into pieces and each piece of the program can be run on a parallel processor. This speeds up the execution of a parallel program. However, certain portion of the parallel program may be inherently sequential and cannot be parallelized. If a program requires 500 milliseconds to run sequentially, calculate the parallel execution time and speed up for the following: (i) 5 processors and 100% of the program can be parallelized. (ii) 5 processors and 70% of the program can be parallelized. Consider a Von Neumann machine with a clock rate of 10 GHz (Trillion Cycles/Second). The average floating point instruction execution requires 8 cycles. Express the computation power of the machine in FLOPS (Floating Point Operations/Second).Explanation / Answer
Ans 4. Parallelization is a core strategic planning. Parallel computers consist of thousands of processors. These computers provide many orders of magnitude. Analyzing the performance of given parallel algorithm calls for comprehensive methods. One such method is Amdahl law.
According to Amdahl law. When a serialized program is parallelized. The execution time of parallelized program is
Tp= Ts.a + Ts .(1-a)/p
Where Ts is the serialized program execution time.
a is fraction of program that cannot be parallelized
p is number of processors
Tp is execution time of parallelized program
According to this,
1) Given Ts= 500 milliseconds
p=5
a =0
Then,
Tp=Ts.a + Ts(a)/p
Tp=500.0+500(1)/5
Tp=100 milliseconds
2) Similarly,
According to Amdahl law
Tp=Ts*a + Ts(1-a)/p
Tp=500*0.3 + 500(0.7)/5
Tp= 200 milliseconds
Ans 5. In Von neumann machine
Computation power = 10 * 109 * 8 = 8 * 10*10= 80 GFLOPS.
Related Questions
drjack9650@gmail.com
Navigate
Integrity-first tutoring: explanations and feedback only — we do not complete graded work. Learn more.