Supported EC2 parameters

IDEA made job submission on EC2 very easy and is fully integrated with EC2. Below is a list of parameters you can specify when you request your simulation to ensure the hardware provisioned will exactly match your simulation requirements.

Compute

base_os

  • Description: Reference to the base OS of the AMI you are using

  • Allowed Values: amazonlinux2 centos7 rhel7

  • Default: If not specified, value default to the OS of the install AMI

  • Examples:

    • -l base_os=centos7: Instances provisioned will be deployed against CentOS manifest

ht_support

Disabled by default

  • Description: Enable support for hyper-threading

  • Allowed Values: yes true no false (case insensitive)

  • Examples:

    • -l ht_support=True: Enable hyper-threading for all instances

    • -l ht_support=False: Disable hyper-threading for all instances (default)

instance_ami

  • Description: Reference to a custom AMI you want to use

  • Default: If not specified, value default to the AMI specified during installation

  • Examples:

    • -l instance_ami=ami-abcde123: Capacity provisioned for the job will use the specific AMI

If you are planning to use an AMI which is not using the same OS as the scheduler, you will need to specify base_os parameter

instance_profile

  • Description: Reference to a custom IAM role to use. Make sure to specify the Instance Profile name and not IAM role name. Refer to the pre-requisites below before using custom IAM role.

  • Default: Use the default IAM instance profile configured by IDEA

  • Examples:

    • -l instance_profile=CustomInstanceProfileName: Compute nodes will a custom IAM Role.

instance_type

  • Description: The type of instance to provision for the simulation

  • Examples:

    • -l instance_type=c5.large: Provision a c5.large for the simulation

    • -l instance_type=c5.large+m5.large: Provision c5.large and m5.large (if needed) for the simulation.

You can specify multiple instances type using "+" sign. When using more than 1 instance type, AWS will prioritize the capacity based on the order (eg: launch c5.large first and switch to m5.large if AWS can't provision c5.large anymore)

nodes

  • Description:The number of EC2 instance to provision

  • Examples:

    • -l nodes=5: Provision 5 EC2 instances

force_ri

  • Description: Restrict a job to run on Reserved Instance

  • Allowed Values: True False

  • Default: False

  • Examples:

    • -l force_ri=False: Job can use RI, On-Demand or Spot

    • -l force_ri=True: Job will only use Reserved Instance. Job will stay in the queue if there is not enough reserved instance available

security_groups

  • Description: Attach additional security groups (max 4) to the compute nodes

  • Allowed Values: sg-xxxxxx

  • Default: False

  • Examples:

    • -l security_groups=sg-abcde: Will attach sg-abcde on top of the default existing security group (ComputeNodeSG)

    • -l security_groups=sg-abcd+sg-efgh: Will attach sg-abcde and sg-efgh on top of the default existing security group (ComputeNodeSG)

You can specify a maximum of 4 additional security groups

spot_allocation_count

  • Description: Specify the number of SPOT instances to launch when provisioning both OD (On Demand) and SPOT instances

  • Allowed Values: Integer

  • Examples:

    • -l nodes=10 -l spot_price=auto -l spot_allocation_count=8: Provision 10 instances, 2 OD and 8 SPOT with max spot price capped to OD price

    • -l nodes=10 -l spot_price=1.4 -l spot_allocation_count=5: Provision 10 instances, 5 OD and 5 SPOT with max spot price set to $1.4

    • -l nodes=10 -l spot_price=auto: Only provision SPOT instances

    • -l nodes=10: Only provision OD instances

This parameter is ignored if spot_price is not specified spot_allocation_count must be lower that the total number of nodes you are requesting (eg: you can not do -l nodes=5 -l spot_allocation_count=15)

spot_allocation_strategy

  • Description: Choose allocation strategy when using multiple SPOT instances type

  • Allowed Valuess: capacity-optimized or lowest-price or diversified (only for SpotFleet deployments)

  • Default Value: capacity-optimized

  • Examples:

    • -l spot_allocation_strategy=capacity-optimized: AWS will provision Spot compute nodes for both EC2 Auto Scaling and EC2 Fleet from the most-available Spot Instance pools by analyzing capacity metrics.

spot_price

  • Description: Enable support for SPOT instances

  • Allowed Values: any float value or auto

  • Examples:

    • -l spot_price=auto: Max price will be capped to the On-Demand price

    • -l spot_price=1.4: Max price you are willing to pay for this instance will be $1.4 an hour.

spot_price is capped to On-Demand price (e.g: Assuming you are provisioning a t3.medium, AWS will default maximum spot price to 0.418 (OD price) even though you specified -l spot_price=15)

subnet_id

  • Description: Reference to a subnet ID to use

  • Default: If not specified, value default to one of the three private subnets created during installation

  • Examples:

    • -l subnet_id=sub-123: Will provision capacity on sub-123 subnet

    • -l subnet_id=sub-123+sub-456+sub-789: + separated list of private subnets. Specifying more than 1 subnet is useful when requesting large number of instances

    • -l subnet_id=2: IDEA will provision capacity in 2 private subnets chosen randomly

If you specify more than 1 subnet and have placement_group set to True, IDEA will automatically provision capacity and placement group on the first subnet from the list

Storage

Refer to the Shared-Storage module if you are looking for more persistent / non-job specify storage backed.

EBS

keep_ebs

Disabled by default

  • Description: Retain or not the EBS disks once the simulation is complete

  • Allowed Values: yes true false no (case insensitive)

  • Default Value: False

  • Example:

    • -l keep_ebs=False: (Default) All EBS disks associated to the job will be deleted

    • -l keep_ebs=True: Retain EBS disks after the simulation has terminated (mostly for debugging/troubleshooting procedures)

root_size

  • Description: Define the size of the local root volume

  • Unit: GB

  • Example: -l root_size=300: Provision a 300 GB SSD disk for / (either sda1 or xvda1)

scratch_size

  • Description: Define the size of the local root volume

  • Unit: GB

  • Example: -l scratch_size=500: Provision a 500 GB SSD disk for /scratch

scratch disk is automatically mounted on all nodes associated to the simulation under /scratch

instance_store

scratch_iops

  • Description: Define the number of provisioned IOPS to allocate for your /scratch device

  • Unit: IOPS

  • Example: -l scratch_iops=3000: Your EBS disks provisioned for /scratch will have 3000 dedicated IOPS

It is recommended to set the IOPs to 3x storage capacity of your EBS disk

FSx for Lustre

fsx_lustre

WITH NO S3 BACKEND

  • Example: -l fsx_lustre=True: Create a new FSx for Lustre and mount it accross all nodes

  • FSx partitions are mounted as /fsx. This can be changed if needed

  • If fsx_lustre_size is not specified, default to 1200 GB

WITH S3 BACKEND

  • Example: -l fsx_lustre=my-bucket-name or -l fsx_lustre=s3://my-bucket-name : Create a new FSx for Lustre and mount it across all nodes

MOUNT EXISTING FSX

  • Description: Mount an existing FSx to all compute nodes if fsx_lustre points to a FSx filesystem's DNS name

  • Example: -l fsx_lustre=fs-xxxx.fsx.region.amazonaws.com

fsx_lustre_size

  • Description: Create an ephemeral FSx for your job and mount the S3 bucket specified

  • Unit: GB

  • Example: -l fsx_lustre_size=3600: Provision a 3.6TB EFS disk

If fsx_lustre_size is not specified, default to 1200 GB (smallest size supported)

This parameter is ignored unless you have specified fsx_lustre=True

fsx_lustre_deployment_type

  • Description: Choose what type of FSx for Lustre you want to deploy

  • Allowed Valuess: SCRATCH_1 SCRATCH_2 PERSISTENT_1 (case insensitive)

  • Default Value: SCRATCH_2

  • Example: -l fsx_lustre_deployment_type=scratch_2: Provision a FSx for Lustre with SCRATCH_2 type

If fsx_lustre_size is not specified, default to 1200 GB (smallest size supported)

This parameter is ignored unless you have specified fsx_lustre=True

fsx_lustre_per_unit_throughput

  • Description: Select the baseline disk throughput available for that file system

  • Allowed Values: 50 100 200

  • Unit: MB/s

  • Example: -l fsx_lustre_per_unit_throughput=250:

Per Unit Throughput is only avaible when using PERSISTENT_1 FSx for Lustre

This parameter is ignored unless you have specified fsx_lustre=True

Network

efa_support

  • Description: Enable EFA support

  • Allowed Values: yes, true, True

  • Example: -l efa_support=True: Deploy an EFA device on all the nodes

You must use an EFA compatible instance, otherwise your job will stay in the queue

ht_support

Disabled by default

  • Description: Enable support for hyper-threading

  • Allowed Values: yes true (case insensitive)

  • Example: -l ht_support=True: Enable hyper-threading for all instances

placement_group

Enabled by default

  • Description: Disable placement group

  • Allowed Values: yes true (case insensitive)

  • Example: -l placement_group=True: Instances will use placement groups

Placement group is enabled by default as long as the number of nodes provisioned is greater than 1

Others

system_metrics

Default to False

  • Description: Send host level metrics to your OpenSearch (formerly Elasticsearch) backend

  • Allowed Values: yes no true false (case insensitive)

  • Example: -l system_metrics=False

Enabling system_metrics generate a lot of data (especially if you are tracking 1000s of nodes). If needed, you can add more storage to your AWS OpenSearch (formerly Elasticsearch) cluster

anonymous_metrics

Default to the value specified during SOCA installation

Last updated