Domain 3: Deployment and Orchestration of ML Workflows

Weight: 22% of scored content

This domain covers deploying models and automating ML pipelines.

Topics Covered

Topic	Description
SageMaker Endpoints	Real-time, Serverless, Async
Batch Transform	Batch inference
SageMaker Pipelines	MLOps workflows
Step Functions	Workflow orchestration
Container Deployment	ECR, ECS, EKS
Auto Scaling	Scaling endpoints

Key Concepts

Deployment Options

graph TD
    A[Model Artifact] --> B{Deployment Type}
    B --> C[Real-time Endpoint]
    B --> D[Serverless Endpoint]
    B --> E[Async Endpoint]
    B --> F[Batch Transform]
    C --> G[Low Latency]
    D --> H[Cost Optimization]
    E --> I[Large Payloads]
    F --> J[Batch Processing]

Choosing Endpoint Type

Type	Latency	Cost	Use Case
Real-time	Milliseconds	Pay per hour	Interactive apps
Serverless	Seconds (cold start)	Pay per request	Variable traffic
Async	Minutes	Pay per hour	Large payloads
Batch	Hours	Pay per job	Offline processing

Study Checklist

• Understand endpoint types and when to use each
• Know SageMaker Pipelines components
• Understand Step Functions for orchestration
• Know container requirements for deployment
• Understand auto-scaling configuration