Sample code to setup NAS Parallel Benchmark using EFA and AWS Batch

This is the sample code for the AWS Batch Blog: Run High Performance Computing Workloads using AWS Batch MultiNode Jobs with Elastic Fabric Adapter

License

This library is licensed under the MIT-0 License. See the LICENSE file.

AWS Batch + EFA

To get started, clone this repo locally:

git clone https://github.com/aws-samples/aws-batch-efa.git
cd aws-batch-efa-blogpost/

AWS Batch Resources

In part 1, we'll create all the necessary AWS Batch resources.

cd batch-resources/

First we'll create a launch template, this launch template installs EFA on the instance and configures the network interface to use EFA. In the launch_template.json file substitute <Account Id>, <Security Group>, <Subnet Id> and <KEY-PAIR-NAME> with your your information.

Now create the launch template:

aws ec2 create-launch-template --cli-input-json file://launch_template.json

To ensure optimal physical locality of instances, we create a placement group, with strategy cluster.

aws ec2 create-placement-group --group-name "efa" --strategy "cluster" --region [your_region]

Next, we'll create the compute environment, this defines the instance type, subnet and IAM role to be used. Edit the <same-subnet-as-in-LaunchTemplate> and <account-id> sections with the pertinent information. Then create the compute environment:

aws batch create-compute-environment --cli-input-json file://compute_environment.json

Finally we need a job queue to point to the compute environment:

aws batch create-job-queue --cli-input-json file://job_queue.json

Dockerfile

In part 2, we build the docker image and upload it to Elastic Container Registery (ECR), so we can use it in our job.

cd ..
pwd # you should be in the aws-batch-efa-blogpost/ directory

First we'll build the docker image, to help with this, we included a Makefile, simply run:

make

Then you can push the docker image to ECR, first modify the top of the Makefile with your region and account id:

AWS_REGION=[your region]
ACCOUNT_ID=[your account id]

Next, push to ECR, note the Makefile assumes you have an ECR repo named aws-batch-efa:

make push     # logs in, tags, and pushes to ECR

Job Definition

cd batch-resources/

Now we need a job definition, this defines which docker image to use for the job, edit the job_definition.json file and substitute <ACCOUNT ID> and <REGION>. Then create the job definition:

aws batch register-job-definition --cli-input-json file://job_definition.json
{
    "jobDefinitionArn": "arn:aws:batch:us-east-1:<account-id>:job-definition/EFA-MPI-JobDefinition:1",
    "jobDefinitionName": "EFA-MPI-JobDefinition",
    "revision": 1
}

Submit a job

Go back to the main directory:

cd ..

Finally we can submit a job!

make submit

Credit

• Arya Hezarkhani, Software Development Engineer, AWS Batch
• Jason Rupard, Principal Systems Development Engineer, AWS Batch
• Sean Smith, Software Development Engineer II, AWS HPC