Run your first job

In this tutorial, follow along to run a program over OpenMPI in a few lines of Python.

Set up

This tutorial runs on Camber infrastructure, and a corresponding notebook is in the demos folder of your Jupyter hub. The easiest way to run this code is to log into Camber and try it out:

1. From your Camber account, click Jupyter Hub to access your Jupyter environment.

2. In the left-hand menu, note the demos folder. It contains subfolders for each demo. For this tutorial, head into demos > 10-get-started > 10-run-your-first-job.

This folder has three files. Open mpi_hello_world.ipynb and start running commands in the notebook.

If you prefer to just read, here’s the full notebook.

Run “Hello, world!” on MPI

This notebook demostrates how to run a simple MPI job on Camber.

1. Use make to build a “Hello, world!” C program.
2. Run the program over MPI.

First, import camber. This instantiates a set of tools to run different types of workloads on Camber Cloud.

import camber

The create_job function creates a process on Camber infrastructure. This function takes at a minimum a command and engine_size as arguments. To compile the C program in the cloud, pass command="make" and engine_size="XXSMALL".

The engine_size specifies the amount of computational power to allocate to the job. Each step up in engine size doubles the allocated compute capacity.

make = camber.mpi.create_job(
    command="make",
    engine_size="XXSMALL"
)

To check the status of the make job, call on its status attribute. When the job finishes, job status will be COMPLETED.

make.status

'COMPLETED'

When the make job is finished, a binary named mpi_hello_world is created in the same directory. This is a compiled C program that prints Hello world from each process.

Run it with MPI with 4 cores and 4 MPI processes:

run = camber.mpi.create_job(
    command="mpirun -np 4 mpi_hello_world",
    engine_size="XXSMALL"
)

Again, check the status of the run job with run.status. When the job finishes, the status returns COMPLETED.

run.status

'COMPLETED'

In fact, you can read the logs of any job with read_logs, which prints the logs of the job to the console:

run.read_logs()

The log should look something like this:

Hello world from processor mpi-master-c6df7-adfie, rank 0 out of 4 processors
Hello world from processor mpi-master-c6df7-adfie, rank 1 out of 4 processors
Hello world from processor mpi-master-c6df7-adfie, rank 2 out of 4 processors
Hello world from processor mpi-master-c6df7-adfie, rank 3 out of 4 processors

You could also download the log file with download_log. This will save the log file to your current working directory under the name job_<JOB_ID>.log.

run.download_log()

Congratulations! You’ve just run your first MPI job on Camber Cloud!