Batch

AWS Batch Overview

A fully managed service for running batch computing workloads on AWS infrastructure (EC2 or ECS/Fargate).

Key Benefits

1. Zero infrastructure management overhead

2. Automatic provisioning and scaling

3. No installation required

Core Components

1. Jobs

• Basic unit of work in AWS Batch

• Can be implemented as:

  • Shell scripts

  • Executable files

  • Docker containers

2. Job Definitions

• Blueprints for job execution

• Specify:

  • Resource requirements

  • Environment configuration

  • Job parameters

3. Job Queues

• Holding area for submitted jobs

• Jobs wait here until:

  • Resources are available

  • Scheduling conditions are met

• Can set multiple priority levels

4. Compute Environments

Types:

1. Managed

   • AWS handles capacity and instance management

   • Automatic scaling

   • ECS/Fargate support

2. Unmanaged

   • User controls compute resources

   • Greater customization options

   • Manual scaling

Here's a visual representation of how these components interact:

Fargate	EC2
Recommended approach for MOST workloads	Need more control over instance selection
Requires fast start times (<30 seconds)	Elastic Fabric Adapter
16 vCPU or less	custom AMIs
no GPUs	GPUs
max 120 GiB of memory	High levels of cocurrency
	access to Linux Parameters

AWS Batch	AWS Lambda
no time limit	15 min execution time
	limited native disk space, and EFS requires functions live within a VPC. It addes overhead and costs.
Batch uses Docker so can have any runtime	fully serverless with limited runtimes

Batch Managed	Batch Unmanage
AWS manages capacity and instance types	managed your own resouces
Compute resource specs are defined when environment is created	AmI must meet ECS AMI specs
ECS instances are launched into VPC subnets	you manage everything
Default is the most recetn and approved Amazon ECS AMI	Good choice for extremely complex or specific requirements
You can use own AMI