Reserved Instances are AWS's primary discount mechanism for compute resources. A 1-year All Upfront Reserved Instance for a c5.xlarge provides roughly 40% savings compared to On-Demand pricing; a 3-year All Upfront provides approximately 60% savings. For workloads running continuously, this is essentially free money — you're going to run these resources anyway, and the commitment just changes when you pay and how much.
The complexity of RIs comes from the specificity of the commitment and the rules around how reservations apply to actual usage. Understanding these rules is the difference between an RI strategy that achieves the expected savings and one that generates waste through poor coverage or mismatched reservations.
RI Purchase Models
Three payment models offer different upfront vs. ongoing cost tradeoffs:
All Upfront: Pay the full reservation cost at purchase time. Provides the largest discount (approximately 40% for 1-year, 60% for 3-year on most instance types). Best for predictable baseline workloads where the capital expense is manageable.
Partial Upfront: Pay a portion upfront and the remainder in monthly installments. Slightly smaller discount than All Upfront. Useful when capital budget is limited but the workload is committed.
No Upfront: No payment at purchase — monthly charges only. Smallest discount of the three options (roughly 30% for 1-year on most types). Useful when committing with no cash outlay is required, though the monthly commitment is contractually binding for the full term.
All three models involve a financial commitment for the term. Unlike Savings Plans, RIs that go unused (because the reserved instance type isn't running) are wasted — you pay for capacity you don't use. This makes purchase accuracy critical.
Standard RIs vs. Convertible RIs
Standard RIs apply to a specific instance family, size, region, OS, and tenancy. You can modify some attributes (instance size within the same family, if instance size flexibility is supported), but the reservation is relatively specific. Provides the full discount for matching usage.
Convertible RIs are more flexible — you can exchange them for different configurations during the term, and they can be converted to different instance families. The trade-off is a smaller discount: roughly 30-40% for 1-year Convertible RIs versus 40% for Standard. The conversion flexibility is valuable if you're uncertain about your future instance mix or expect the workload to change significantly during the term.
For established, stable workloads with predictable instance type usage, Standard RIs provide better economics. For evolving workloads or where there's uncertainty about future instance type needs, Convertible RIs reduce the risk of ending up with unusable capacity.
Instance Size Flexibility
For Linux/Unix standard RIs in the EC2 default tenancy, instance size flexibility allows a single RI to apply to any size within the same instance family in the same region. An m5.2xlarge RI (16 vCPUs, 32 GB RAM) can apply to two m5.xlarge instances, or four m5.large instances, or one m5.xlarge plus two m5.medium instances — any combination that consumes the equivalent normalized units.
Normalization factors: nano=0.25, micro=0.5, small=1, medium=2, large=4, xlarge=8, 2xlarge=16, 4xlarge=32, 8xlarge=64, 16xlarge=128, 32xlarge=256. A reservation for one m5.2xlarge (16 units) covers the same capacity as four m5.large instances (4 units × 4 = 16 units). This flexibility makes it reasonable to buy larger RIs that provide flexibility across your instance size mix.
Using Cost Explorer's RI Recommendations
The AWS Cost Explorer RI Recommendations tool analyzes your actual On-Demand usage history and recommends specific RI purchases to maximize savings. The tool accounts for your current RI coverage, provides projected savings, and shows the payback period for each recommendation.
Review recommendations with skepticism about the coverage percentage they recommend. The tool's goal is maximizing savings, which can lead to recommendations for 100% coverage of a usage pattern that's actually more variable than 14 or 30 days of history suggest. Cover 70-80% of your consistent baseline with RIs, leaving on-demand capacity for growth and variability. This approach slightly reduces maximum savings but eliminates the risk of stranded capacity from over-purchasing.
RDS Reserved Instances
RDS RIs work similarly to EC2 RIs but apply to specific database engines, instance types, and deployment configurations (single-AZ vs. multi-AZ). Unlike EC2 RIs, RDS RIs don't have instance size flexibility — a db.r5.large RI only applies to db.r5.large usage, not other sizes in the r5 family.
The specificity of RDS RIs makes purchase accuracy more important. Analyze at least 30 days of RDS On-Demand usage before purchasing RDS RIs to confirm the database type and size. RDS instances tend to be more stable than EC2 (databases aren't scaled down to zero when not needed), making them good RI candidates.
Related Reading
- Savings Plans vs Reserved Instances — when to use each discount model
- AWS cost optimization guide — RIs in the context of overall cost management
- Spot Instances guide — complementing RIs with spot capacity for variable workloads
- Billing anomaly detection — monitoring that RI coverage is performing as expected
FAQ
What happens to unused Reserved Instances?
Unused RI capacity is wasted — you pay the reservation cost whether the capacity is used or not. If you have RIs that consistently don't match running instances, you can sell Standard RIs on the AWS Marketplace Reserved Instance Marketplace, transferring the reservation to another buyer. Convertible RIs can't be sold on the marketplace but can be exchanged for different configurations. Selling unused RIs typically recovers most (but not all) of the remaining reservation value.
Can I modify Reserved Instances after purchase?
Standard RIs can be modified to split into smaller sizes (within the same family and region), merge smaller sizes into larger, or change the Availability Zone. Convertible RIs can be exchanged for a new Convertible RI of equal or greater value in a different configuration. Payment terms (All Upfront, Partial Upfront, No Upfront) cannot be changed after purchase.
Should I buy 1-year or 3-year RIs?
1-year RIs provide sufficient savings (typically 35-40%) while limiting commitment to a manageable timeframe. 3-year RIs offer larger discounts (55-60%) but require high confidence in the workload's stability. A common approach is 1-year All Upfront for stable production workloads, with some 3-year coverage for your most stable baseline infrastructure (core databases, always-on services).
Protect your AWS accounts before it's too late
Vigilare monitors your AWS accounts for suspension risks — billing anomalies, IAM issues, GuardDuty findings, and more — and alerts you before AWS takes action.
Written by Vigilare Engineering
Platform Team