I/O, Debugging, and Performance¶

I/O¶

Debugging¶

Comm.Barrier()¶

Blocks until all processes in the communicator have reached this routine

Parameters: Comm (MPI comm) – communicator on which we are to block processes

Example:

Performance: Speedup and Efficiency¶

Wtime() and Wtick()¶

Example:

#timeExample.py
from mpi4py import MPI
import numpy
comm = MPI.COMM_WORLD
rank = comm.Get_rank()

comm.Barrier()
start = MPI.Wtime()

A = numpy.array([[1.,2.,3.],[4.,5.,6.],[7.,8.,9.]])
local_a = numpy.zeros(3)
comm.Scatter(A,local_a)
print "process", rank, "has", local_a

comm.Barrier()
end = MPI.Wtime()

print end - start

Amdahl’s Law¶

Exercises¶

Some examples may display super-linear speedup, at least at some point
Do a speedup analysis of the trap rule…