Concept and mechanism
Optimization starts with understanding usage and removing waste with evidence. A Compute Savings Plan corresponds to an eligible usage commitment over a term; it should not automatically be sized to peak demand or treated as capacity reservation. Spot can serve interruptible work when checkpoints and resumption are appropriate. Checkpoints need to survive the instance. A task tolerating delay and repetition has different options from an operation requiring continuity without recovery mechanisms.
Guided application
Network costs require observing the actual path: cross-AZ traffic to a zonal NAT gateway may incur charges absent from a simple gateway count. Compare alternatives by volume, fixed cost, and resilience; an S3 gateway endpoint may remove eligible traffic from the NAT path. Budgets with alerts inform owners but do not create an absolute billing cap. When retiring applications, inventory volumes and snapshots, confirm retention and recovery, identify owners, and delete only unnecessary resources. Retain the decision so remaining expenses can be explained.
The December peak does not represent the annual baseline and part of the application will retire. First estimate remaining usage before proposing a commitment.
Common pitfalls
Buying discounts before removing waste; keeping checkpoints only locally; deleting volumes without ownership or a retention decision.
Related topics: Identity, trust, and permissions · Networks, endpoints, and hybrid connectivity
Assess total cost against the same service and data requirements.
Reference: Savings Plans overview · SAA-C03