View a markdown version of this page

Networking modes for Amazon Batch jobs - Amazon Batch
Services or capabilities described in Amazon Web Services documentation might vary by Region. To see the differences applicable to the China Regions, see Getting Started with Amazon Web Services in China (PDF).

Networking modes for Amazon Batch jobs

The following table describes the networking modes and typical usage for Amazon Batch job types. See the links in the "Job type" column for more details regarding considerations and behaviors.

Job Type Supported Network Mode(s) Typical Usage
ECS-EC2 simple job host Used for the highest scalable embarrassingly parallel batch workloads that only require egress to vpc defined in Compute Environment.
ECS-EC2 multi-node parallel job awsvpc Used for tightly-coupled, multi-host (node) distributed workloads modeled as a single job with coordinated communications between task nodes.
ECS-Fargate simple job awsvpc True serverless for embarrassingly parallel batch workloads. Typically lowest TCO and highest container isolation job model.
Amazon ECS Managed Instances simple job host Fully managed Amazon EC2 instances with broader compute flexibility than Fargate (GPU, bare metal, arbitrary vCPU/memory). Uses host networking for higher network throughput.
EKS-EC2 simple job host and pod Used for highly scalable embarrassingly parallel batch workloads that only require egress to vpc defined in Compute Environment. Default is host networking.
EKS-EC2 multi-node parallel job host and pod Used for tightly-coupled, multi-host (node) distributed workloads modeled as a single job with coordinated communications between pod nodes. Default is host networking.