Management and Governance
AWS Auto Scaling
AWS Auto Scaling coordinates scaling plans and predictive or dynamic scaling for supported resources. In self-managed GenAI stacks it aligns capacity with workload signals across services while respecting configured bounds.
Key points
- Target tracking maintains a metric near a chosen value
- Predictive scaling prepares capacity from historical patterns
- Minimum, maximum, cooldown, and warm-up settings constrain reactions
When to use it
- Scale a custom inference fleet on request concurrency
- Adjust several supporting resources for a scheduled batch window
Exam tips
- Choose service-native autoscaling when one resource is involved and AWS Auto Scaling for coordinated plans
- Use token throughput, queue age, or accelerator memory signals when CPU hides model saturation