19. Jupyter Notebook Setup¶

To use Bodo with Jupyter Notebook, install jupyter and ipyparallel in your Bodo Conda environment:

conda install jupyter ipyparallel mpi4py -c conda-forge

Create an MPI profile for IPython:

ipython profile create --parallel --profile=mpi

Edit the ~/.ipython/profile_mpi/ipython_config.py file and add the following line:

c.IPClusterEngines.engine_launcher_class = 'MPIEngineSetLauncher'

Start the Jupyter notebook in your Bodo Conda environment:

jupyter notebook

Go to IPython Clusters tab and select the number of engines (i.e., cores) you’d like to use and click Start next to the mpi profile. Alternatively, you can use ipcluster start -n 4 --profile=mpi in a terminal to start the engines. Initialization of the engines can take several seconds.

Now start a new notebook and run this code in a cell to setup the environment:

import ipyparallel as ipp
c = ipp.Client(profile='mpi')
view = c[:]
view.activate()
view.block = True  # equivalent to running with %%px --block

# Set the working directory:
import os
view["cwd"] = os.getcwd()
%px cd $cwd

This should complete without any errors. An error may appear if the cluster is not initialized yet (usually NoEnginesRegistered). In this case, just wait a few seconds and try again.

You can now run Bodo functions on the execution engines using ipyparallel hooks such as %%px magic and the work will be distributed across the engines. For example, run this code in a cell:

%%px
import bodo
import numpy as np
import time

@bodo.jit
def calc_pi(n):
    t1 = time.time()
    x = 2 * np.random.ranf(n) - 1
    y = 2 * np.random.ranf(n) - 1
    pi = 4 * np.sum(x**2 + y**2 < 1) / n
    print("Execution time:", time.time()-t1, "\nresult:", pi)

calc_pi(2 * 10**8)

If you wish to run across multiple nodes, you can add the following to ipcluster_config.py:

c.MPILauncher.mpi_args = ["-machinefile", "path_to_file/machinefile"]

where machinefile (or hostfile) is a file with the hostnames of available nodes that MPI can use. More information about machinefiles can be found here.

It is important to note that other MPI systems and launchers (such as QSUB/PBS) may use a different user interface for the allocation of computational nodes.