The Jeff’s Note #
While preparing for the AWS SAA-C03, many candidates confuse Reserved Instances with Capacity Reservations. In the real world, this is fundamentally a decision about temporal commitment vs. capacity guarantee granularity. Let’s drill into a simulated scenario.
The Scenario #
GlobalTech Summit Inc. is organizing a 7-day international developer conference in the US-East region. The event will host live-streamed workshops, real-time code compilation environments, and interactive AI demos. To ensure seamless participant experience, the infrastructure team needs guaranteed compute capacity for 150 c5.4xlarge instances distributed across three specific Availability Zones (us-east-1a, us-east-1b, us-east-1c) to achieve fault tolerance.
The event runs for exactly one week, after which all resources will be terminated. The finance team has flagged that unused commitment beyond this period is unacceptable.
Key Requirements #
Secure guaranteed EC2 capacity in three specific Availability Zones for a 1-week duration with no long-term financial commitment.
The Options #
- A) Purchase Reserved Instances specifying the required region.
- B) Create On-Demand Capacity Reservations specifying the required region.
- C) Purchase Reserved Instances specifying the required region and three Availability Zones.
- D) Create On-Demand Capacity Reservations specifying the required region and three Availability Zones.
Correct Answer #
Option D.
The Architect’s Analysis #
Correct Answer #
Option D - Create On-Demand Capacity Reservations specifying the required region and three Availability Zones.
Step-by-Step Winning Logic #
-
Capacity Guarantee Precision: ODCRs can be scoped to specific AZs, ensuring capacity is reserved exactly where your architecture requires it (us-east-1a/b/c). Regional reservations don’t guarantee AZ-level availability.
-
No Temporal Lock-In: Unlike RIs (1-3 year commitments), ODCRs are pay-as-you-go. You reserve capacity for the event duration, launch instances against the reservation, and cancel immediately after—paying only for the 7 days used.
-
Billing Alignment: ODCRs charge On-Demand rates whether instances run or not, but you avoid the 51-75% waste of a 1-year RI commitment for a 1-week event.
-
Zonal Architecture Match: The scenario explicitly states “three specific Availability Zones”—this is a signal that AZ-level scoping is architecturally required (likely for latency, data residency, or fault isolation).
The Traps (Distractor Analysis) #
-
Why not A (Regional RI)?
Reserved Instances are billing discounts, not capacity guarantees. Even with regional scope, AWS doesn’t guarantee capacity availability during high-demand periods. You also commit to 1+ years of payment for a 7-day event. -
Why not B (Regional ODCR)?
Regional ODCRs provide capacity guarantees but don’t allow AZ-level specification. AWS fulfills the reservation in any AZ within the region. If us-east-1d gets your capacity instead of 1a/1b/1c, your architecture breaks. -
Why not C (Zonal RI)?
While Zonal RIs do provide AZ-specific capacity reservation as a secondary benefit, the 1-year minimum commitment violates the FinOps requirement. You’d pay for 52 weeks to use 1 week—a 5100% cost overhead.
The Architect Blueprint #
Diagram Note: Three zonal ODCRs ensure capacity is reserved in architecturally required AZs, with Auto Scaling Groups launching instances against these reservations during the event window.
The Decision Matrix #
| Option | Commitment | AZ Control | Est. Cost (150 instances, 7 days) | Pros | Cons |
|---|---|---|---|---|---|
| A: Regional RI | 1-3 years | ❌ None | Billing discount (if long-term) | No capacity guarantee; massive overcommitment | |
| B: Regional ODCR | None | ❌ Region-only | ~$16,800 (On-Demand rate) | No long-term commitment | Cannot specify required AZs; architectural risk |
| C: Zonal RI | 1-3 years | ✅ AZ-specific | Capacity + AZ guarantee | 5100% cost overhead for 1-week need | |
| D: Zonal ODCR ✅ | None | ✅ AZ-specific | ~$16,800 (On-Demand rate, 7 days only) | Precise capacity + AZ control; zero waste | Slightly higher per-hour rate vs. RI (irrelevant at 7 days) |
Cost Calculation Basis: c5.4xlarge On-Demand ~$0.68/hr × 150 instances × 168 hours = $17,136. RI 1-year All Upfront ~$0.42/hr equivalent, but 52-week commitment = $109,368 total.
Real-World Practitioner Insight #
Exam Rule #
For the SAA-C03 exam:
- Reserved Instances = Long-term cost optimization (1-3 years).
- On-Demand Capacity Reservations = Short-term capacity guarantees without commitment.
- Look for keywords: “specific AZ” → Zonal ODCR; “1-week”, “temporary event” → avoid RIs.
Real World #
In production, we’d layer additional strategies:
-
Scheduled Reservations (if still available in your region): AWS deprecated these in 2021, but they were ideal for recurring events.
-
Savings Plans: For recurring annual events, we’d use Compute Savings Plans (1-year) with baseline capacity, supplementing peaks with ODCRs.
-
Capacity Blocks for ML (if AI workloads): For GPU instances, Capacity Blocks offer fixed-duration reservations (1-8 hours) at discounted rates.
-
Hybrid Commitment: Reserve 70% of expected capacity via 1-year Convertible RIs (if events recur quarterly), using ODCRs only for the 30% surge buffer.
-
Pre-warming Communication: Engage AWS account team 2-4 weeks before large events. While ODCRs guarantee capacity, proactive coordination helps with bulk IP allocation and service limit increases.