MPI Training

Version v0.2 of the documentation is no longer actively maintained. The site that you are currently viewing is an archived snapshot. For up-to-date documentation, see the latest version.

This guide will walk you through using MPI for training

Installing MPI Operator

If you haven’t already done so please follow the Getting Started Guide to deploy Kubeflow.

An alpha version of MPI support was introduced with Kubeflow 0.2.0. You must be using a version of Kubeflow newer than 0.2.0.

Verify that MPI support is included in your Kubeflow deployment

Check that the MPI Job custom resource is installed

kubectl get crd

The output should include

NAME                                       AGE
...                       4d

If it is not included you can add it as follows

ks pkg install kubeflow/mpi-job
ks generate mpi-operator mpi-operator
ks apply ${ENVIRONMENT} -c mpi-operator

Creating an MPI Job

You can create an MPI Job by defining an MPIJob config file. See Tensorflow benchmark example config file. You may change the config file based on your requirements.

cat examples/tensorflow-benchmarks.yaml

Deploy the MPIJob resource to start training:

kubectl create -f examples/tensorflow-benchmarks.yaml

You should now be able to see the created pods matching the specified number of GPUs.

kubectl get pods -l mpi_job_name=tensorflow-benchmarks-16

Training should run for 100 steps and takes a few minutes on a gpu cluster. Logs can be inspected to see its training progress.

PODNAME=$(kubectl get pods -l mpi_job_name=tensorflow-benchmarks-16,mpi_role_type=launcher -o name)
kubectl logs -f ${PODNAME}

Monitoring an MPI Job

kubectl get -o yaml mpijobs tensorflow-benchmarks-16

See the status section to monitor the job status. Here is sample output when the job is successfully completed.

kind: MPIJob
  clusterName: ""
  creationTimestamp: 2018-08-14T19:48:44Z
  generation: 1
  name: tensorflow-benchmarks-16
  namespace: default
  resourceVersion: "7670207"
  selfLink: /apis/
  uid: 0d24b791-9ffb-11e8-9b38-029ed2ab0d38
  gpus: 16
      creationTimestamp: null
      - image: mpioperator/tensorflow-benchmarks:latest
        name: tensorflow-benchmarks
        resources: {}
  launcherStatus: Succeeded