Homework 10:  Due March 22, 2007

Use the program I gave you to write your own barrier routine in homework 6 to time new barrier
routines that you will write.  Time them exactly as was done in homework 6.  The following are the
barrier routines you are supposed to write:

1.  subroutine barrier1(p, comm)          ! write a barrier routine using mpi_bcast, as was done
for hw6.

2.  subroutine barrier2(p, rank, comm) ! write a barrier routine using mpi_sendrecv or
mpi_sendrecv_replace
                                                              ! this will count 5 points.

3.  subroutine barrier3(p, comm)          ! write a barrier routine using mpi_alltoall
                                                              ! this will count 5 points

4. subroutine barrier4(p, rank, comm)  ! write a barrier routine using mpi_isend~Rs and
mpi_irecv~Rs.
                                                              ! this will count 5 points

5. subroutine barrier5(p, rank, comm)  ! write a barrier routine using processor 0 as a central
manager, i.e.
                                                              ! all nonzero processors send a
message to processor 0 when they reach
                                                              ! the barrier and then processor 0
sends a message to all other processors
                                                              ! when all processors have reached
the barrier.
                                                              ! this will count 5 points

A single program will compare the timings of the 5 barriers above with mpi_barrier using bmpirun
with 4, 8, and 16 processors.  Each run will give 1 timing trial for each barrier.   Make two runs
for 4 processors, two runs for 8 processors, and two runs with 16 processors.   The timing portion
will be worth 5 points.