What's changed: Service coverage: added a catalog section (s4) for developer tools and operations management (CI/CD, Grafana/Prometheus, Proton, Amplify/Device Farm)
3.2Performance and Cost Optimization
Understand optimizing existing workloads—right-sizing, purchase options (On-Demand/RI/Savings Plans/Spot), Compute Optimizer/Cost Explorer/Budgets/Trusted Advisor, and CloudFront/caching. Keep performance while cutting waste.
Cost optimization rests on right-sizing and choosing purchase options. Boost performance with caching and content delivery.
3.2.1Cost and performance optimization
- Purchase options: steady = Savings Plans/RI (commit for discount), variable = On-Demand, interruption-OK batch = Spot (max discount).
- Right-sizing: Compute Optimizer detects over/under-provisioned resources and recommends sizes.
- Cost visibility: Cost Explorer (analysis), Budgets (alerts), Trusted Advisor (optimization checks).
- Performance: CloudFront (CDN) for edge delivery, ElastiCache to offload the DB and cut latency.
Common on SAP-C02: discount for always-on = Savings Plans/RI, cheapest interruption-tolerant batch = Spot, sizing recommendations = Compute Optimizer, spend analysis = Cost Explorer / budget alerts = Budgets, delivery acceleration = CloudFront. Use commitment plans for steady load, Spot for burst batch.
Spot can be interrupted, so it suits stateless, checkpointable workloads. A classic mix covers steady always-on with Savings Plans and absorbs overflow with On-Demand.
SAP-C02 probes the fine distinctions between purchase options. Savings Plans commit to an hourly spend ($/h) for 1–3 years: Compute Savings Plans apply automatically across instance family, region, and OS—and even Fargate/Lambda—while EC2 Instance Savings Plans are limited to a family but offer a deeper discount. Reserved Instances (RIs) split into zonal RIs (with a capacity reservation) and flexible regional RIs (no capacity guarantee). Spot gives a 2-minute interruption notice and can pre-empt evictions via capacity rebalancing; an EC2 Auto Scaling mixed-instances policy blends On-Demand/RI/Spot to balance availability and cost. On the visibility side, Cost Explorer surfaces RI/Savings Plans purchase recommendations plus coverage/utilization, Budgets alerts on cost, usage, or RI utilization, and Trusted Advisor runs cross-cutting checks for idle and underutilized resources.
| Demand pattern | Best purchase option | Why |
|---|---|---|
| Steady for 1–3 yrs | Savings Plans / RI | Deepest discount via commitment |
| Short-term / unpredictable | On-Demand | No commitment, instant flex |
| Interruption-OK batch/CI | Spot | Max discount, 2-min interruption notice |
| Dedicated for licensing/compliance | Dedicated Hosts | Dedicated physical host, BYOL |
Scenario: A steady always-on web tier coexists with large nightly video encoding (interruption-OK). Minimize cost. → Cover the web tier with Compute Savings Plans so it follows family changes; run encoding on Spot in an Auto Scaling group with a mixed-instances policy; set a monthly cap alert with Budgets; and periodically right-size the web tier with Compute Optimizer.
FAQ: Savings Plans or RIs? For flexibility, Compute Savings Plans (auto-apply across family/region/OS, covering Fargate/Lambda). If you need a capacity reservation in a specific family, use a zonal RI. Discounts are broadly comparable, and Savings Plans are generally recommended today.
Exam trap: Choosing Spot to cheapen a "stateful, can-not-be-interrupted production database" is wrong—an interruption can lose state. Steady workloads call for Savings Plans/RI; Spot assumes interruption tolerance.
3.2.2Section summary
- Purchase = Savings Plans/RI (steady) / On-Demand (variable) / Spot (interruptible)
- Visibility = Cost Explorer/Budgets/Trusted Advisor; performance = CloudFront/ElastiCache
Sign in to track progress — Log in.
Quick check
(just a quick review)Q1. You want maximum savings on compute for a production workload running steadily for 3 years. What?
Q2. You want to run large interruption-tolerant batch jobs at the lowest possible cost. What?
Q3. You want analysis of whether EC2/EBS are over-provisioned, with right-size recommendations. What?