Skip to main content
GullySystem
A practical guide

Reserved Instances vs Savings Plans vs On-Demand: When to Commit

Ganesh HSStrategy and Technology, GullySystem
Practical advice · 3 min read

In this guide 7 sections
The short answer

Commit only after right-sizing, and only on the baseline that runs all year. AWS describes Savings Plans as a commitment to hourly spend, with more flexibility than Reserved Instances. Variable load stays on demand. GullySystem’s cloud cost review keeps commitment coverage below full usage.

Three ways to pay for the same machine

On-demand means paying for what runs, with no commitment. It has the highest hourly rate and allows the most change. It suits workloads that are new, short-lived or still being sized, because nothing is lost when they change shape, move region or stop altogether.

Reserved Instances and Savings Plans both trade a one-year or three-year commitment for a lower rate. The AWS Savings Plans FAQ lists Compute, EC2 Instance, Database and SageMaker plans.

Azure and Google Cloud offer their own versions under other names. The question of when to commit is the same.

How Reserved Instances and Savings Plans differ

A Reserved Instance is tied to a configuration such as instance type and region. If the workload moves, the discount may need an exchange or modification to follow it.

A Savings Plan commits you to a set spend per hour. AWS says Compute Savings Plans apply across instance types and regions. That suits a setup still changing shape.

The AWS FAQ suggests moving from Reserved Instances to Savings Plans as they expire. Read the current terms for your workload before buying either.

Why commitment comes after right-sizing

Committing first locks in machines that were due to shrink. The discount then applies to capacity you never needed. That is the trap.

The safer order is to remove idle resources, schedule test environments, tier storage and right-size. Then watch the settled usage for a while before committing.

  • Idle resources removed
  • Non-production environments on a schedule
  • Instances sized from measured use
  • Several weeks of settled usage observed

How much to commit

Cover the floor, not the peak. The floor is the usage that runs every hour of every day. Anything above it stays on demand.

Full coverage looks efficient on a spreadsheet. It turns into a burden when the architecture changes or a workload moves to serverless. A business that commits to its entire current usage has little room left when a product is retired, a database is resized or traffic moves to another region.

Stagger purchases where you can. Commitments ending at different times leave room to adjust.

Who should hold the decision

Commitments are a finance call as much as a technical one. The payment options run from all upfront to no upfront, which affects cash flow.

Keep commitments in your own cloud account. The discount then shows on your own bill, where you can verify it. Reseller arrangements make that harder.

  • Term length against product plans
  • Upfront payment against cash position
  • Coverage level against the measured floor
  • Expiry dates in the finance calendar

Commitment planning sheet

A sheet listing each steady workload with its measured hourly floor, its expected life and any planned change. Add a column for the commitment type that fits. Workloads with a planned change in the coming year stay on demand.

Open a blank worksheet to print

Questions owners ask

Should we take a three-year term for the lower rate?

Not unless the workload is settled. A longer term locks in the current design. If a migration or redesign is planned, a shorter term or on-demand keeps options open.

Can a Savings Plan be cancelled?

Generally no. The AWS FAQ describes a narrow return window for small, recent purchases. Treat a commitment as fixed for its full term.

Do commitments cut the bill without other work?

Not on their own. They lower the rate, not the waste. An idle or oversized machine under a commitment is still idle or oversized.

Sources

Browse all guidesBack to top ↑
Next step

Have a specific situation to work through?

This article covers the general case. Tell us what you’re actually dealing with and we’ll respond directly.

See the Cloud Cost Optimisation service