Parallel Reader Tutorial

Setup a parallel jupyter notebook using ipyparallel

See http://ipyparallel.readthedocs.io for more details on how to set this up



In [1]:

    
import ipyparallel

# attach to a running cluster
cluster = ipyparallel.Client(profile='mpi')
print('profile:', cluster.profile)
print("IDs:", cluster.ids) # Print process id numbers









    



('profile:', u'mpi')
('IDs:', [0, 1, 2, 3])






    



/usr/local/lib/python2.7/dist-packages/ipyparallel/client/client.py:458: RuntimeWarning: 
            Controller appears to be listening on localhost, but not on this machine.
            If this is true, you should specify Client(...,sshserver='you@uzi')
            or instruct your controller to listen on an external IP.
  RuntimeWarning)

Create a serial reader

Every serial reader extends the BaseReader class.
Should be able to read an arbitrary sub domain of the full grid
Ability to set and get the viz dump, domain size, etc.



In [2]:

    
%%px

from mpi4py import MPI
import numpy

import matplotlib
%matplotlib inline
import matplotlib.pyplot as plt

comm = MPI.COMM_WORLD
rank = comm.Get_rank()

print "rank = %d" %rank

# Import the WCHR Ascii reader module
import floatpy.readers.wchr_ascii_reader as war

# Set filename prefix for the dataset (this is specific to this serial reader)
filename_prefix = '/home/akshays/Data/WCHR-regent/2017_AIAA_AVIATION/WCHR/CHIT2_'

# Initialize the serial reader based on your data format
serial_reader = war.WchrAsciiReader(filename_prefix)

print "Domain size: ", serial_reader.domain_size









    



[stdout:0] 
rank = 0
Domain size:  (64, 64, 64)
[stdout:1] 
rank = 1
Domain size:  (64, 64, 64)
[stdout:2] 
rank = 2
Domain size:  (64, 64, 64)
[stdout:3] 
rank = 3
Domain size:  (64, 64, 64)

Create the parallel data reader

The ParallelDataReader class takes any concrete implementation of the abstract BaseReader
Parallelization is built in automatically
3D decomposed data
Supports ghost cell communication and transpose operations



In [3]:

    
%%px

import floatpy.readers.parallel_reader as pdr

reader = pdr.ParallelDataReader( MPI.COMM_WORLD, serial_reader, num_ghosts=(0, 0, 0) )

print "rank %d: Sub domain: " %(comm.Get_rank()), reader.serial_reader.sub_domain









    



[stdout:0] rank 0: Sub domain:  ((0, 0, 0), (31, 31, 63))
[stdout:1] rank 1: Sub domain:  ((32, 0, 0), (63, 31, 63))
[stdout:2] rank 2: Sub domain:  ((0, 32, 0), (31, 63, 63))
[stdout:3] rank 3: Sub domain:  ((32, 32, 0), (63, 63, 63))

Read in coordinates and data



In [4]:

    
%%px

# Read the coordinates
x, y, z = reader.readCoordinates()



In [5]:

    
%%px

# Set the visualization step
reader.step = 200

# Read in velocity data for this viz dump
u, v, w = reader.readData( ('u', 'v', 'w') )

Plotting at runtime

The ParallelPlane class gathers the full plane data given a normal and index



In [10]:

    
%%px

import floatpy.utilities.parallel_plane as pp

# Create an object to extract an x-y plane with z index 32
plane = pp.ParallelPlane( reader.grid_partition, direction=2, index=0 )

# Get x and y coordinates on the plane
has_plane, x_p = plane.get_plane(x)
has_plane, y_p = plane.get_plane(y)

# Get u velocity on the plane
has_plane, u_p = plane.get_plane(u)

if has_plane:
    plt.figure(figsize=(10,6))
    plt.imshow( u_p, extent=[x_p[0,0], x_p[-1,-1], y_p[0,0], y_p[-1,-1]], origin='lower', interpolation='lanczos', aspect=1. )
    plt.colorbar()
    plt.show()









    





[output:0]

Parallel compact derivatives



In [14]:

    
%%px

from floatpy.derivatives.compact import CompactDerivative


dx = x[1,0,0] - x[0,0,0]
dy = y[0,1,0] - y[0,0,0]
dz = z[0,0,1] - z[0,0,0]

der = CompactDerivative(reader.grid_partition, (dx, dy, dz), (6, 6, 6), reader.periodic_dimensions)



In [16]:

    
%%px

vorticity = der.curl(u, v, w)

has_plane, vz_p = plane.get_plane(vorticity[:,:,:,2])

if has_plane:
    plt.figure(figsize=(10,6))
    plt.imshow( vz_p, extent=[x_p[0,0], x_p[-1,-1], y_p[0,0], y_p[-1,-1]], origin='lower', interpolation='lanczos', aspect=1. )
    plt.colorbar()
    plt.show()









    





[output:0]



In [ ]: