Fall 2006 CS 431/632 Parallel Computing
Homework-3
10/24/2006 1-2
Individual work only. 200 points. Due Nov 7, 2006.
1. Implement the collective communication primitive “broadcast” (one-to-all) using point-to-
point message passing primitives provided by MPI using the following algorithms:
a. linear
b. ring
c. binary tree
For each of the above algorithms plot the time taken by the broadcast function and the
speedup curve for different number of processes. Note that you cannot use any MPI collective
function calls to implement the broadcast function. Use a single driver program to test
different algorithms. The driver program should test each broadcast function for different
messages sizes also (use a loop inside the driver program to vary the message size from 25 to
220).
Use the following table to include the timing measurements for each algorithm:
Processes
Bytes
1 2 4 6 8 16 32 64
32
64
128
256
512
1024
2048
4098
8192
16384
32768
65536
131072
262144
524288
1048576
2. If tm = tsend = trecv is the time taken for sending/receiving a message between any two
processes develop expressions to compute the total time spent in communication (tcomm) for
each of the above algorithms. [Graduate Students Only]
