While preparing for the AWS SAP-C02, many candidates get confused by mixing workload criticality with instance purchasing models. In the real world, this is fundamentally a decision about SLA protection vs. cost optimization through intelligent capacity planning. Let’s drill into a simulated scenario.
The Scenario #
GlobalAnalytics Inc. operates a business intelligence platform currently hosted in their corporate data center using 12 fully-redundant servers configured for high availability. The system executes two distinct workload types:
- Scheduled Analytics Jobs: Hourly and daily batch processes consuming 65% of total compute capacity. These jobs have strict SLA requirements and typical execution windows ranging from 20 minutes to 2 hours.
- Ad-Hoc User Queries: On-demand analytical requests initiated by business analysts, consuming 35% of capacity. These tasks typically complete within 5 minutes and have no SLA guarantees.
During infrastructure failures, the business mandate is clear: scheduled jobs must continue meeting SLA commitments, while user-initiated tasks may experience delays.
The CTO has mandated migration to AWS EC2 with the following non-negotiable constraints:
- Pay-as-you-go pricing with no long-term commitments
- Maintain high availability equivalent to current on-premises setup
- Preserve SLA compliance for scheduled workloads
- Minimize total cost of ownership
Key Requirements #
Design an EC2 deployment architecture that balances workload criticality, availability requirements, and cost optimization using appropriate instance purchasing models and multi-AZ distribution.
The Options #
- A) Deploy 12 instances across two Availability Zones; each AZ runs 2 On-Demand instances with Capacity Reservations + 4 Spot instances.
- B) Deploy 12 instances across three Availability Zones; one AZ runs 4 On-Demand instances with Capacity Reservations, remaining instances are Spot.
- C) Deploy 12 instances across three Availability Zones; each AZ runs 2 On-Demand instances with Savings Plans + 2 Spot instances.
- D) Deploy 12 instances across three Availability Zones; each AZ runs 3 On-Demand instances with Capacity Reservations + 1 Spot instance.
Correct Answer #
Option A.
The Architect’s Analysis #
Correct Answer #
Option A — Deploy across two Availability Zones with 2 On-Demand + 4 Spot instances per AZ.
Step-by-Step Winning Logic #
This solution demonstrates professional-grade capacity planning by aligning purchasing models with workload characteristics:
1. Workload-to-Purchasing Model Mapping #
- 4 On-Demand instances with Capacity Reservations (33% of fleet) provide guaranteed capacity for the 65% workload share through resource overcommitment — acceptable because:
- Scheduled jobs have predictable execution windows
- Not all 12 instances run at 100% simultaneously
- SLA jobs get priority scheduling on guaranteed capacity
- 8 Spot instances (67% of fleet) handle the 35% best-effort workload at 70-90% cost savings
2. Multi-AZ Strategy Aligned with HA Requirements #
- Two AZs provide sufficient fault tolerance for the stated “high availability” requirement without over-engineering
- Each AZ maintains identical capacity mix (2 OD + 4 Spot), enabling active-active workload distribution
- AZ failure scenario: Remaining AZ’s 2 On-Demand instances can absorb critical SLA workload while Spot handles overflow
3. FinOps Compliance with “No Long-Term Commitment” #
- Capacity Reservations are pay-as-you-go (charged whether used or not, but no 1-3 year lock-in)
- Avoids Savings Plans (Option C) which require commitment despite flexibility
- Spot instances maintain cost discipline on non-critical workload
The Traps (Distractor Analysis) #
Why not Option B? #
- Single point of concentration risk: Placing all 4 On-Demand instances in one AZ creates an availability bottleneck
- Imbalanced failure resilience: Loss of the primary AZ immediately violates SLA (only Spot remains in other AZs)
- Over-reliance on Spot for critical workload: 8 Spot instances across 2 AZs means SLA jobs compete with best-effort tasks during Spot reclamation events
Why not Option C? #
- Violates “no long-term commitment” constraint: Savings Plans require 1 or 3-year commitments
- Insufficient guaranteed capacity: Only 6 On-Demand instances (50% of fleet) creates SLA risk during peak scheduled job windows
- Three AZs unnecessary: Adds 33% cost overhead for AZ data transfer and complexity without proportional availability gain
Why not Option D? #
- Over-provisioning of guaranteed capacity: 9 On-Demand instances (75% of fleet) for a 65% SLA workload wastes capital
- Insufficient Spot allocation: Only 3 Spot instances (25% of fleet) for 35% best-effort workload forces costly On-Demand usage for user queries
- Poor cost optimization: Fails to maximize Spot savings opportunity
The Architect Blueprint #
Diagram Note: SLA-critical scheduled jobs route primarily to Capacity-Reserved On-Demand instances (orange) with Spot overflow capacity, while best-effort user queries leverage cost-optimized Spot instances (green) across two Availability Zones.
The Decision Matrix #
| Option | Est. Complexity | Est. Monthly Cost | Pros | Cons |
|---|---|---|---|---|
| A (Correct) | Medium | $3,200 (4 OD m5.2xlarge CapRes @ $0.384/hr = $1,105 8 Spot @ ~$0.12/hr = $691 + Data transfer ~$400) |
✅ Perfect workload-to-cost alignment ✅ SLA protection via guaranteed capacity ✅ 78% Spot cost savings on 67% of fleet ✅ Balanced multi-AZ resilience |
⚠️ Requires intelligent job scheduler ⚠️ Spot interruption handling needed |
| B | Low | $2,950 (4 OD CapRes = $1,105 8 Spot = $691 3 AZ transfer = $350) |
✅ Lowest operational complexity ✅ Three-AZ distribution |
❌ Single AZ concentration risk ❌ SLA violation on primary AZ failure ❌ Imbalanced capacity distribution |
| C | High | $4,100 (6 OD with 1yr SP @ $0.277/hr = $1,990 6 Spot = $518 3 AZ transfer = $450) |
✅ Three-AZ distribution ✅ Lower OD hourly rate with SP |
❌ Violates “no commitment” requirement ❌ Insufficient guaranteed capacity (50%) ❌ Higher total cost despite SP discount ❌ 3-AZ overhead unnecessary |
| D | Medium | $5,200 (9 OD CapRes = $2,488 3 Spot = $259 3 AZ transfer = $450) |
✅ Maximum SLA safety margin ✅ Three-AZ resilience |
❌ 62% cost premium over Option A ❌ Over-provisioned guaranteed capacity ❌ Underutilized Spot savings opportunity ❌ Poor FinOps efficiency |
Cost Assumptions: m5.2xlarge instances (8 vCPU, 32 GB RAM) in us-east-1; On-Demand $0.384/hr, Spot average $0.115/hr (~70% savings); 730 hours/month; data transfer estimated at 5TB/month cross-AZ.
Real-World Practitioner Insight #
Exam Rule #
For SAP-C02, when you see “mixed workload criticality + no long-term commitment + cost optimization”, the answer pattern is:
- Match purchasing model to workload SLA (On-Demand/Reserved for critical, Spot for best-effort)
- Capacity Reservations ≠ long-term commitment (they’re pay-as-you-go)
- Avoid over-engineering multi-AZ (2 AZs often sufficient unless explicitly stated otherwise)
Real World #
In production, I would enhance Option A with:
- Spot Fleet diversification: Use multiple instance types (m5, m5a, m5n) across diversified Spot pools to reduce interruption probability from 5-10% to <2%
- Auto Scaling + Predictive Scaling: Adjust On-Demand baseline during known peak windows (month-end reporting) using CloudWatch metrics
- Savings Plans consideration after 3 months: Once workload patterns stabilize, evaluate Compute Savings Plans (no instance family lock-in) to reduce the On-Demand portion by additional 15-20% while maintaining flexibility
- Hybrid Reserved Instances: For the absolute baseline (e.g., 2 instances always running), consider Convertible RIs after 6 months of stable operations
- Third AZ for true mission-critical: If SLA penalties exceed $10K/hour, the additional ~$400/month for three-AZ deployment becomes insurance, not overhead
The exam tests purchasing model understanding; reality requires continuous FinOps optimization based on observed workload telemetry.