AWS SAA-C03Free AWS Certified Solutions Architect — Associate practice test

Correct answer: C. Transition to S3 Standard-IA on day 8, expire on day 90.

S3 Standard-IA costs roughly $0.0125/GB/month versus Standard's $0.023/GB/month, making it the right tier for the 'rarely accessed' window between days 8 and 90. Objects are never accessed after day 90, so a Lifecycle expiration on day 90 deletes them immediately — no need to pay for a Glacier transition for data that will be immediately deleted. S3 Standard-IA has a 30-day minimum storage duration charge, but transitioning on day 8 still saves money compared to remaining in Standard for the full 90-day period. Option A skips Standard-IA for the days 8–90 window where data is merely rarely (not never) accessed and applies Glacier Instant Retrieval unnecessarily early; Glacier retrieval fees would apply to any rare access event. Option B delays the Standard-IA transition to day 30, wasting 22 days of Standard pricing, and delays deletion to day 180, charging for 90 extra days of Glacier storage. Option D uses Intelligent-Tiering, whose per-object monitoring fee ($0.0025 per 1,000 objects/month) makes it less optimal when the access pattern is already fully known.

Correct answer: B. Purchase 1-year Standard Reserved Instances with an all-upfront payment for the baseline capacity.

Standard Reserved Instances with all-upfront payment for a 1-year term provide the deepest discount for a specific instance family, OS, and region — up to ~40% over On-Demand for a 1-year term, reaching up to ~60% for 3-year terms. EC2 Instance Savings Plans match Standard RI discounts but are not listed as an option here. The all-upfront payment variant maximizes effective savings within the 1-year commitment tier by eliminating financing charges. Option A (Spot) is unsuitable for a critical stable baseline because Spot Instances can be interrupted with a two-minute warning. Option C (Compute Savings Plans) offers up to ~66% savings but applies compute-family flexibility rather than a specific instance type commitment; for a fully predictable, stable workload using a known instance type the Standard RI typically provides equal or greater savings than a Compute Savings Plan for the same term. Option D (3-year Convertible no-upfront) provides a meaningful discount but the no-upfront payment dilutes the effective per-hour savings compared to all-upfront, and Convertible RIs carry a lower discount ceiling (~20% less) than Standard RIs.

Correct answer: B. Spot Instances launched via an EC2 Auto Scaling group with a mixed-instances policy.

Spot Instances offer up to 90% savings over On-Demand and are ideal for fault-tolerant, interruptible workloads. Using a mixed-instances Auto Scaling group spreads demand across multiple Spot capacity pools, reducing the likelihood of simultaneous interruptions and allowing the nightly batch to complete reliably. Option A uses On-Demand which is the most expensive per-hour option and offers no discount for intermittent use. Option C (Reserved Instances) charges the same per-hour rate regardless of whether instances are running — buying RI capacity for only 4 hours/night means paying for 20 idle hours out of every 24, making it far more expensive than On-Demand for this usage pattern. Option D (Dedicated Hosts) is designed for software licensing compliance requirements, not cost reduction, and is significantly more expensive for this use case.

Correct answer: C. S3 Glacier Deep Archive with a Lifecycle expiration rule after 7 years.

S3 Glacier Deep Archive is the lowest-cost AWS storage class at approximately $0.00099/GB/month and supports retrieval within 12 hours using the standard retrieval option, which satisfies the audit SLA. The 7-year Lifecycle expiration rule automates deletion at end of retention. Option B (Glacier Flexible Retrieval) also meets the 12-hour SLA (standard retrieval tier) but costs roughly $0.004/GB/month — approximately 4× more than Deep Archive for the same infrequent access pattern. Option A (Standard-IA) costs around $0.0125/GB/month, far more expensive than Glacier tiers for data that is almost never accessed. Option D (EFS IA) is a network file system designed for active workloads shared across EC2 instances; at approximately $0.025/GB/month for Infrequent Access storage it is the most expensive option listed and inappropriate for cold archival.

Correct answer: C. AWS Compute Optimizer

AWS Compute Optimizer uses machine learning to analyze CloudWatch utilization metrics (CPU, memory, network, disk I/O) and recommends optimal EC2 instance types, Auto Scaling group configurations, EBS volume types, and Lambda memory sizes. Option A (Trusted Advisor) provides basic cost checks such as flagging idle or low-utilization EC2 instances, but it does not generate ML-driven right-sizing recommendations with specific alternative instance suggestions across families. Option B (CUR) is a detailed billing data export; it surfaces cost and usage data but has no analytical or recommendation capability. Option D (AWS Config) records and evaluates resource configuration compliance and change history but has no cost-optimization or right-sizing functionality.

Correct answer: A. The traffic is routing over a NAT Gateway; replace it with a VPC Gateway Endpoint for S3.

When EC2 instances in a VPC access S3 through a NAT Gateway, all traffic is billed at the NAT Gateway data-processing rate (~$0.045/GB) plus NAT Gateway hourly fees, even though S3 is in the same region. A VPC Gateway Endpoint for S3 routes traffic directly over the private AWS network at no data-processing charge, eliminating the NAT Gateway cost entirely for S3-bound traffic. Option B (S3 Transfer Acceleration) is an opt-in feature for faster cross-internet uploads; it is not automatically enabled and charges appear as a separate line item only when explicitly activated. Option C (consolidating AZs) would reduce cross-AZ data transfer charges for inter-instance traffic, but it does not explain or address NAT Gateway charges on S3-bound traffic. Option D is incorrect because S3 does not charge data-transfer fees for EC2-to-S3 traffic within the same region; the cost driver here is the NAT Gateway processing fee, not S3 egress.

Correct answer: B. Add an ElastiCache for Redis cluster in front of RDS to cache frequent query results.

ElastiCache (Redis or Memcached) caches the results of frequent, repetitive read queries in memory, so those queries never reach RDS at all. This directly reduces the number of RDS I/O operations and therefore the associated I/O cost. Option C (Read Replicas) offloads read traffic from the primary but still generates I/O on each replica — the application pays for I/O on both the primary (for replication writes) and the replica (for read queries), so overall I/O billing may actually increase. Option A (Aurora MySQL) uses a distributed log-structured storage engine that can reduce I/O charges significantly, but it requires a full database engine migration — a more complex, higher-risk change compared to adding a caching layer. Option D (Provisioned IOPS io2) increases I/O throughput and reduces latency but does not reduce the number of I/O operations billed; it would raise storage costs rather than lower them.

Correct answer: C. AWS Lambda with no provisioned concurrency, paying only per-invocation.

Lambda's default on-demand pricing charges only for actual invocations and compute duration — there is zero cost when there are no invocations. This perfectly matches an unpredictable, bursty API with zero-traffic days. Option B (provisioned concurrency) reserves pre-initialized execution environments that are billed continuously by the hour even when completely idle, adding unnecessary fixed cost for a workload that has zero-traffic days. Option A (EC2 Spot) incurs per-hour instance charges even when the Auto Scaling group is at minimum capacity; a minimum instance count still accrues charges during zero-traffic periods. Option D (Fargate with minimum 1 task) pays for at least one running container 24/7 at vCPU + memory rates, which creates idle resource charges on days with no traffic.

Correct answer: A. Provisioned capacity with Auto Scaling targeting 70% utilization, min 50 RCU / 10 WCU, max 2,500 RCU / 600 WCU.

Provisioned capacity with Auto Scaling is cheapest when traffic is low for most of the month and spikes are brief and predictable. At baseline Auto Scaling holds capacity near 50 RCU / 10 WCU (very low cost); it scales to 2,500 RCU / 600 WCU only for the ~15-minute spike window. On-Demand pricing charges per request unit and can be approximately 6–7× more expensive than provisioned for sustained low-throughput usage because you pay a premium per RCU/WCU for the operational flexibility. Option B (On-Demand) is ideal when traffic is truly unpredictable or difficult to forecast; with a known and brief monthly spike, provisioned + Auto Scaling is materially cheaper over the full month. Option C permanently provisions 2,000 RCU / 500 WCU for the entire month to cover only 15 minutes of need — an enormous waste. Option D adds DAX cluster costs (per-node hourly billing, minimum one node) for a read acceleration layer that is not justified by 15-minute monthly spikes that Auto Scaling handles adequately.

Correct answer: B. Migrate volumes to EBS gp3 and set provisioned IOPS to 3,000 and throughput to 125 MB/s.

gp3 volumes cost $0.08/GB/month compared to gp2 at $0.10/GB/month — a 20% reduction — and gp3 allows independent provisioning of IOPS and throughput. Given the very low queue depth (0.1) and throughput well under 50 MB/s, the gp3 baseline specification of 3,000 IOPS and 125 MB/s throughput is more than sufficient for these analytics workloads and no additional IOPS need to be purchased above baseline. Option A (io2 Block Express) is the highest-cost EBS type at $0.125/GB/month plus $0.065 per provisioned IOPS; it is designed for I/O-intensive databases requiring sub-millisecond latency — completely inappropriate for already-underutilized analytics volumes. Option C (st1 HDD) is lower cost per GB ($0.045/GB/month) but st1 is HDD-based, cannot serve as a boot volume, and delivers poor latency for random I/O patterns common in analytics workloads; a significant performance regression is likely. Option D (snapshotting and resizing) does not reduce per-GB storage costs and introduces operational risk from data deletion.