Pass Your Amazon AWS Certified Solutions Architect - Associate SAA-C03 Exam Easy!

When demand is stable, well-used provisioned capacity and supported auto scaling can be cheaper; neither model is universally best.

The provisioned option can be economical for predictable demand, but static capacity sized to a trough may throttle at a higher known peak. Auto scaling and a sufficient minimum are needed to match the measured hourly demand.

Standard-IA decreases storage cost but increases the price of requests relative to Standard. Table class is not the same as provisioned-versus-on-demand throughput mode and is not automatically cheaper for request-heavy payments.

The Standard-IA class has different storage and throughput pricing; it fits high storage-to-request ratios, not all tables.

Standard can be preferable when reads and writes dominate cost, but this audit table is mainly charged for retaining many terabytes of seldom-requested data. Compare storage plus all requests, not request rates alone.

TTL is a deletion feature, not a lower-cost class for keeping required data. Deleting retained audit records to reduce storage can breach stated compliance needs; Standard-IA retains the records with different storage and request pricing.

The attribute name alone does not activate TTL processing. The table must have TTL enabled on the designated Number-typed Unix epoch seconds attribute, and deletion is asynchronous rather than instantaneous.

Streams record item changes for consumers; they do not schedule expiration-based deletion merely because an item has an expiresAt field. Native TTL provides managed eventual deletion when configured on the table.

A Scan filter excludes items from a returned result after reads. It does not delete data, so expired records continue occupying storage and consuming reads until a separate delete or the TTL background process removes them.

EC2 Instance Plans apply to EC2 family/Region usage, not managed RDS; eligible RDS RI or Database Savings Plans are separate instruments.

RDS reserved-instance discounts are subject to eligible Region, engine, licensing, instance and deployment matching rules. A mismatched commitment can leave the actual PostgreSQL instance charged at On-Demand rates.

A reservation is a billing discount matched to eligible running usage, not a generic discount that applies to any unrelated DB class, engine or Region. Adding an unneeded replica would create an additional billable instance.

Provisioned capacity incurs compute charges while running, even if application traffic is very low, unlike eligible elastic capacity patterns.

RDS storage autoscaling grows allocated disk space when it is running low; it does not vary the provisioned instance class or pause compute. Aurora Serverless v2 uses an ACU range to vary database compute independently.

RDS Proxy can reuse client database connections, but the provisioned Aurora writer still incurs its instance charge while running. Serverless v2 changes the compute capacity model, subject to engine support and possible auto-pause conditions.

Supported RDS stop has a maximum continuous duration and AWS can automatically restart eligible stopped instances.

A manual snapshot supports a restore into a new DB instance but is not a stopped, immediately restartable original instance. Deleting and restoring can introduce endpoint and recovery work; supported stop/start preserves the instance settings.

A stopped RDS instance can avoid eligible compute charges, but allocated storage, provisioned IOPS and backup storage can still be billed. Supported instances also restart automatically after seven consecutive days.

Read replicas are separately running database resources; adding them without load pressure may create cost without useful benefit.

An endpoint alone cannot offload reads or reduce provisioned instance cost. New replicas must be justified by measured load and used by the application, and each running replica has its own cost.

A traditional RDS Multi-AZ DB-instance standby supports failover rather than serving user SELECT requests, and availability capacity is not free. It is not a justified cost-reduction substitute for evaluating actual reader demand.

I/O-Optimized can help I/O-heavy workloads but may be more expensive than Aurora Standard when I/O usage is modest.

Aurora Standard separately bills eligible I/O operations. Ignoring the very I/O charges causing the issue understates Standard total cost relative to the different storage and compute pricing of Aurora I/O-Optimized.

Reducing compute capacity does not switch an Aurora Standard cluster to I/O-Optimized, so Standard I/O metering still applies. Compare storage configuration plus compute and the actual read/write I/O bill.

Larger compute may temporarily add throughput but does not remove repeated work, whereas cache hits can avoid duplicate database reads.

Reader instances only offload eligible SQL when clients use the correct reader endpoints. Keeping all connections on the writer does not eliminate repeated result computation or create a managed cache for identical queries.

RDS Proxy reuses backend connections and can reduce connection churn; it is not an SQL result cache. ElastiCache with a safe cache-aside design can avoid repeated database reads when application freshness permits it.

Each replica contains billed table/index data and receives replicated change traffic with associated write/transfer costs.

DNS answer selection has no authority to waive DynamoDB replication charges.

Global-table replicas are independently billable active DynamoDB tables: they store the indexed data and consume read and replicated-write usage in their Regions. Their managed replication is not a free read-only cache service.

Preallocating speculative capacity increases current storage charges; eligible RDS storage autoscaling can grow up to a defined maximum.

Backup retention governs recovery-history storage, not allocated database volume size. A large live RDS allocation remains billable until an actual supported migration/resizing action changes it; autoscaling avoids future over-allocation at provisioning time.

Changing EBS-backed storage type can affect IOPS and throughput economics, but it does not reclaim previously allocated GiB. RDS storage autoscaling grows when needed and does not automatically shrink an oversized existing volume.