Amazon SAA-C03: Architecture Scenarios and Trade-Offs

SAA-C03 scenarios are usually trade-off problems disguised as service questions. The candidate is given availability, security, performance, operational, and cost requirements and must choose the architecture that satisfies all important constraints with the least unnecessary complexity. Several answer choices may be technically possible; the strongest one is the best fit for the complete requirement.

The examples below stay within the current Amazon SAA-C03 scope. They are not claims about live exam questions. Use them to practice requirement-first reasoning rather than memorized service pairings.

Scenario one: a private application needs S3 without internet or NAT

Instances in private subnets need to access Amazon S3, and the design should avoid sending that traffic through a NAT gateway. A VPC endpoint is stronger than making the instances public or paying for unnecessary egress through NAT.

The key is recognizing that private AWS-service connectivity can satisfy both security and cost goals. The VPC design should follow the traffic path rather than assume every private subnet needs NAT for every dependency.

Scenario two: an image-processing workload arrives in bursts

The workload is event-driven, runs for short periods, and can process each object independently. Lambda can be a strong fit because it scales with events and removes server management. A permanently running EC2 fleet may add unnecessary cost and operations.

If processing duration, memory, specialized runtime, or persistent state exceeds serverless constraints, EC2 or containers may become stronger. The EC2 versus Lambda decision is workload-specific.

Scenario three: a web tier must survive instance and AZ failure

Use stateless application instances across multiple Availability Zones behind a load balancer, with Auto Scaling replacing failed capacity. A single large instance with a standby script provides weaker fault isolation and slower recovery.

The design should also externalize state. High availability at the compute layer is incomplete if local sessions or files tie users to one instance.

Scenario four: database reads are growing but writes remain moderate

A read replica can offload read traffic from the primary relational database. Multi-AZ alone improves availability but does not primarily exist to scale reads. The requirement should decide which replication pattern is needed.

Review RDS Multi-AZ and read replicas until availability and read scaling are clearly separated. Similar words such as “replica” can hide different architectural goals.

Scenario five: static content must load quickly worldwide

CloudFront can cache content at edge locations and reduce origin load. Route 53 can direct users to endpoints but does not itself cache website objects. The two services may be used together, but they solve different layers of the problem.

A CloudFront design is especially strong when objects are cacheable and users are geographically distributed. Origin security and cache invalidation still need consideration.

Scenario six: application components fail together because of synchronous calls

Insert a queue or event-driven boundary so that temporary downstream failure does not block upstream work. SQS can buffer jobs while consumers recover or scale.

However, asynchronous processing introduces retries, duplicate delivery considerations, visibility timeouts, and dead-letter handling. Decoupling improves resilience only when those behaviors are designed explicitly.

Scenario seven: unknown object-access patterns make storage tiering difficult

S3 Intelligent-Tiering can move objects among access tiers based on observed access patterns, reducing the need to predict exactly when each object becomes cold. An explicit lifecycle can still be stronger when age and retention patterns are known.

The choice between Intelligent-Tiering and fixed lifecycle transitions follows predictability, access pattern, and policy requirements rather than one universal cost rule.

Scenario eight: an application needs low-latency key-value access at very large scale

DynamoDB may fit better than a relational database when access patterns are key-oriented and horizontal scale is central. A relational engine is stronger when joins, complex transactions, or relational constraints dominate.

The DynamoDB and NoSQL model should be chosen because it matches the data access pattern, not simply because the workload is large.

Scenario nine: security teams require encryption with controlled key permissions

Use encryption at rest and choose the key-management model that satisfies control, rotation, separation of duties, and audit requirements. Customer-managed KMS keys can provide more control than service-managed keys but also add management responsibility.

The S3 encryption example is useful because the architecture decision includes both data encryption and who may use the key. Encryption without correct key policy can still fail the access requirement.

Scenario ten: the cheapest design would violate the required recovery target

Cost optimization does not mean removing redundancy until the system becomes fragile. If the business requires recovery from an AZ failure with minimal interruption, a single-AZ database or one-instance architecture may be cheaper but unacceptable.

The broader SAA-C03 architecture discipline is to optimize within constraints. First satisfy the business requirement, then remove unnecessary cost. That is the logic behind many exam trade-offs.

Scenario eleven: a company needs a predictable private network connection from its data center, but the connection must also survive a single path failure. One Direct Connect link is not sufficient resilience by itself. The design may require redundant connections or a VPN backup depending on throughput, recovery, and cost. The right answer follows the availability requirement, not simply the desire for dedicated connectivity.

Scenario twelve: a workload processes millions of independent jobs and can tolerate interruption. Spot capacity can reduce compute cost substantially, especially when Auto Scaling or managed batch patterns replace interrupted workers. If the workload is stateful or cannot retry safely, Spot may create unacceptable risk. Cost benefit depends on interruption tolerance.

Scenario thirteen: an internal API experiences unpredictable spikes and the team wants to minimize idle infrastructure. API Gateway and Lambda can provide elastic request handling if the runtime and integration fit. If requests are long-running or need specialized persistent compute, containers or EC2 may be stronger. The architecture should reflect the execution model.

Scenario fourteen: a workload has a large relational database but only a few tables receive heavy reads. Caching or read replicas may be more targeted than scaling the primary vertically. The candidate should identify whether the bottleneck is repeated reads, complex queries, or write throughput before selecting the remedy.

Scenario fifteen: a company wants to retain backups for years but almost never restores them. Use lower-cost archival storage with a restore time that still meets compliance and recovery expectations. Keeping every backup in a hot tier may be operationally simple but unnecessarily expensive.

Scenario sixteen: one microservice failure causes upstream requests to time out. Introduce an asynchronous boundary, circuit-breaking behavior at the application layer, or retries appropriate to the dependency. The architecture should avoid turning a local failure into a system-wide cascade.

Scenario seventeen: a static website stores objects in S3 but the bucket must not be public. Use CloudFront with an appropriate private-origin access pattern rather than exposing the bucket directly. This preserves edge delivery while keeping the storage resource private. The design separates public content delivery from public storage permission.

Scenario eighteen: a company needs to connect hundreds of VPCs. Full-mesh peering becomes difficult to manage because peering is non-transitive and connection count grows quickly. Transit Gateway can provide a more scalable hub-style network when the requirements fit. The architect should choose topology based on scale and routing simplicity.

Scenario nineteen: a workload needs shared POSIX-style files across many Linux instances. EFS is generally a stronger fit than storing the same files on one EBS volume, because EBS is block storage attached under different constraints. The decision follows access semantics, not the word “storage.”

Scenario twenty: a batch job can be retried safely and runs only a few hours each night. Spot Instances or a managed batch architecture can reduce cost if interruption is tolerated. If completion time is rigid and no checkpointing exists, On-Demand or reserved capacity may be more appropriate. Purchasing model is an application-design decision.

Scenario twenty-one: a globally distributed application must keep data in one Region for compliance but needs fast reads worldwide. Edge caching or a read-through content layer may improve user latency without replicating the regulated system of record. Data residency constraints can change which global techniques are acceptable.

Scenario twenty-two: a company wants the lowest operational burden for a relational database but also needs automatic failover inside a Region. A managed RDS or Aurora Multi-AZ design is stronger than self-managing database replication on EC2 unless the scenario requires operating-system control. The trade-off favors managed availability because the business requirement is database service, not custom database administration.

Scenario twenty-three: a workload sends large volumes from private subnets to S3 through NAT gateways. Moving S3 traffic to a gateway endpoint can reduce unnecessary NAT processing and keep the path private. This is a good example of one topology change improving security posture and recurring cost at the same time.

Scenario twenty-four: an application must tolerate a complete Region failure but can accept several minutes of recovery and some operational complexity. A warm-standby or pilot-light style architecture may be more economical than fully active-active multi-Region service. The correct pattern follows the RTO, RPO, and budget rather than assuming the most redundant architecture is always best.

Scenario twenty-five: a team wants to expose a private database directly to the internet because application servers need access from another VPC. Prefer private networking such as peering, Transit Gateway, or another suitable private connectivity pattern rather than weakening the database boundary. Connectivity requirements should be solved at the network architecture layer before public exposure is considered.

Private connectivity preserves a much stronger default security posture.

For every scenario, identify the non-negotiable requirements first, then rank the desirable ones. A design that is cheaper but violates security or recovery requirements is not “more optimized.” The exam rewards candidates who preserve the whole requirement while avoiding unnecessary services and complexity.