Creating a compute node group in AWS PCS - AWS PCS
PrerequisitesCreate a compute node group in AWS PCS

Creating a compute node group in AWS PCS

This topic provides an overview of available options and describes what to consider when you create a compute node group in AWS Parallel Computing Service (AWS PCS). If this is your first time creating a compute node group in AWS PCS, we recommend you follow the tutorial in Get started with AWS Parallel Computing Service. The tutorial can help you create a working HPC system without expanding into all the available options and system architectures that are possible.

Prerequisites

  • Sufficient service quotas to launch the desired number of EC2 instances in your AWS Region. You can use the AWS Management Console to check and request increases to your service quotas.

  • An existing VPC and subnet(s) that meet AWS PCS networking requirements. We recommend that you thoroughly understand these requirements before you deploy a cluster for production use. For more information, see AWS PCS VPC and subnet requirements and considerations. You can also use a CloudFormation template to create a VPC and subnets. AWS provides an HPC recipe for the CloudFormation template. For more information, see aws-hpc-recipes on GitHub.

  • An IAM instance profile with permissions to call the AWS PCS RegisterComputeNodeGroupInstance API action and access to any other AWS resources required for your node group instances. For more information, see IAM instance profiles for AWS Parallel Computing Service.

  • A launch template for your node group instances. For more information, see Using Amazon EC2 launch templates with AWS PCS.

  • To create a compute node group that uses Amazon EC2 Spot instances, you must have the AWSServiceRoleForEC2Spot service-linked role in your AWS account. For more information, see Amazon EC2 Spot role for AWS PCS.

Create a compute node group in AWS PCS

You can create a compute node group using the AWS Management Console or the AWS CLI.

AWS Management Console
To create your compute node group using the console

  1. Open the AWS PCS console.

  2. Select the cluster where you want to create a compute node group. Navigate to Compute node groups and choose Create.

  3. In the Compute node group setup section, provide a name for your node group. The name can only contain case-sensitive alphanumeric characters and hyphens. It must start with an alphabetic character and can't be longer than 25 characters. The name must be unique within the cluster.

  4. Under Computing configuration, enter or select these values:

    1. EC2 launch template – Select a custom launch template to use for this node group. Launch templates can be used to customize network settings such as subnet, and security groups, monitoring configuration, and instance-level storage. If you don't have a launch template prepared, see Using Amazon EC2 launch templates with AWS PCS to learn how to create one.

      Important

      AWS PCS creates a managed launch template for each compute node group. These are named pcs-identifier-do-not-delete. Don't select these when you create or update a compute node group, or the node group won't function correctly.

    2. EC2 launch template version – You must select a version of your custom launch template. If you change the version later, you must update the compute node group to detect changes in the launch template. For more information, see Updating an AWS PCS compute node group.

    3. AMI ID – if your launch template doesn't include an AMI ID, or if you want to override the value in the launch template, provide an AMI ID here. Note that the AMI used for the node group must be compatible with AWS PCS. You can also select a sample AMI provided by AWS. For more information on this topic, see Amazon Machine Images (AMIs) for AWS PCS.

    4. IAM instance profile – Choose an instance profile for the node group. An instance profile grants the instance permissions to access AWS resources and services securely. If you don't have one prepared, see IAM instance profiles for AWS Parallel Computing Service to learn how to create one.

    5. Subnets – Choose one or more subnets in the VPC where your AWS PCS cluster is deployed. If you select multiple subnets, EFA communications won't be available between nodes, and communication between nodes in different subnets might have increased latency. Make sure the subnets you specify here match any that you define in the EC2 launch template.

    6. Instances – Choose one or more instance types to fulfill scaling requests in the node group. All instance types must have the same processor architecture (x86_64 or arm64) and number of vCPUs. If the instances have GPUs, all instance types must have the same number of GPUs.

    7. Scaling configuration – Specify the minimum and maximum number of instances for the node group. You can define either a static configuration, where there is a fixed number of nodes running, or a dynamic configuration, where up to the maximum count of nodes can run. For a static configuration, set minimum and maximum to the same, greater than zero number. For a dynamic configuration, set minimum instances to zero and maximum instances to a number greater than zero. AWS PCS doesn't support compute node groups with a mix of static and dynamic instances.

  5. (Optional) Under Additional settings, specify the following:

    1. Purchase option – select between Spot and On-demand instances.

    2. Allocation strategy – if you have selected the Spot purchase option, you can specify how Spot capacity pools are chosen when launching instances in the node group. For more information, see Allocation strategies for Spot Instances in the Amazon Elastic Compute Cloud User Guide. This option has no effect if you have selected the On-demand purchase option.

  6. (Optional) In the Slurm custom settings section, provide these values:

    1. Weight – This value sets the priority of nodes in the group for scheduling purposes. Nodes with lower weights have higher priority, and the units are arbitrary. For more information, see Weight in the Slurm documentation.

    2. Real memory – This value sets the size (in GB) of real memory on nodes in the node group. It is meant to be used in conjunction with the CR_CPU_Memory option in the Cluster Slurm configuration in AWS PCS. For more information, see RealMemory in the Slurm documentation.

  7. (Optional) Under Tags, add any tags to your compute node group.

  8. Choose Create compute node group. The Status field shows Creating while AWS PCS provisions the node group. This can take several minutes.

Recommended next step

  • Add your node group to a queue in AWS PCS to enable it to process jobs.

AWS CLI
To create your compute node group using AWS CLI

Create your queue with the command that follows. Before running the command, make the following replacements:

  1. Replace regionwith the ID of the AWS Region to create your cluster in, such as us-east-1.

  2. Replace my-cluster with the name or clusterIdof your cluster.

  3. Replace my-node-groupwith the name for your compute node group. The name can contain only alphanumeric characters (case-sensitive) and hyphens. It must start with an alphabetic character and can't be longer than 25 characters. The name must be unique within the cluster.

  4. Replace subnet-ExampleID1 with one or more subnets IDs from your cluster VPC.

  5. Replace lt-ExampleID1 with the ID for your custom launch template. If you don't have one prepared, see Using Amazon EC2 launch templates with AWS PCS to learn how to create one.

    Important

    AWS PCS creates a managed launch template for each compute node group. These are named pcs-identifier-do-not-delete. Don't select these when you create or update a compute node group, or the node group won't function correctly.

  6. Replace launch-template-version with a specific launch template version. AWS PCS associates your node group with that specific version of the launch template.

  7. Replace arn:InstanceProfilewith the ARN of your IAM instance profile. If you don't have one prepared, see Using Amazon EC2 launch templates with AWS PCS for guidance.

  8. Replace min-instances and max-instances with integer values. You can define either a static configuration, where there is a fixed number of nodes running, or a dynamic configuration, where up to the maximum count of nodes can run. For a static configuration, set minimum and maximum to the same, greater than zero number. For a dynamic configuration, set minimum instances to zero and maximum instances to a number greater than zero. AWS PCS doesn't support compute node groups with a mix of static and dynamic instances.

  9. Replace t3.large with another instance type. You can add more instance types by specifying a list of instanceType settings. For example, --instance-configs instanceType=c6i.16xlarge,instanceType=c6a.16xlarge. All instance types must have the same processor architecture (x86_64 or arm64) and number of vCPUs. If the instances have GPUs, all instance types must have the same number of GPUs.

aws pcs create-compute-node-group --region region \
    --cluster-identifier my-cluster \
    --compute-node-group-name my-node-group \
    --subnet-ids subnet-ExampleID1 \
    --custom-launch-template id=lt-ExampleID1,version='launch-template-version' \
    --iam-instance-profile-arn=arn:InstanceProfile \
    --scaling-config minInstanceCount=min-instances,maxInstanceCount=max-instance \
    --instance-configs instanceType=t3.large

There are several optional configuration settings you can add to the create-compute-node-group command.

  • You can specify --amiId if your custom launch template doesn't include a reference to an AMI, or if you wish to override that value. Note that the AMI used for the node group must be compatible with AWS PCS. You can also select a sample AMI provided by AWS. For more information on this topic, see Amazon Machine Images (AMIs) for AWS PCS.

  • You can select between on-demand (ONDEMAND) and Spot (SPOT) instances using --purchase-option. On-demand is the default. If you choose Spot instances, you can also use --allocation-strategy to define how AWS PCS chooses Spot capacity pools when it launches instances in the node group. For more information, see Allocation strategies for Spot Instances in the Amazon Elastic Compute Cloud User Guide.

  • It is possible to provide Slurm configuration options for the nodes in the node group using --slurm-configuration. You can set the weight (scheduling priority) and real memory. Nodes with lower weights have higher priority, and the units are arbitrary. For more information, see Weight in the Slurm documentation. Real memory sets the size (in GB) of real memory on nodes in the node group. It is meant to be used in conjunction with the CR_CPU_Memory option for the cluster in AWS PCS in your Slurm configuration. For more information, see RealMemory in the Slurm documentation.

Important

It can take several minutes to create the compute node group.

You can query the status of your node group with the following command. You won’t be able to associate the node group with a queue until its status reaches ACTIVE.

aws pcs get-compute-node-group --region region \
    --cluster-identifier my-cluster \
    --compute-node-group-identifier my-node-group