QUIZ 2

1. In a computation involving iterative nearest neighbour mesh updates, how does the communication cost vary as the surface to volume ratio of the processor geometry is increased ?

2. What is an MPI Communicator? Why is it necessary to make a collective call, say, to set up communicatiors in Cannon's matrix multiply algorithm ?

3.What is scaled speedup ? Why is it required ?

4.Define scalability. What is an isoefficiency function ?

5. Identify the loop carried dependencies in the following code segment. Write a code segment that fixes this problem. You are given one array a[N] on a CREW shared memory, with the number of available processors = P = N

/* Initialization */

for(i=0;i<N;i++)

a[i] = 3*i + 5*i*i;

/* Iterative computation */

for(t=1;t<NT;t++)

forall(i=1;i<N-1;i++)

a[i] = a[i-1] + a[i+1];