Database capacity modes and reserved capacity: SAP-C03 study notes and practice questions
What to know about Database capacity modes and reserved capacity for the SAP-C03 exam, from the Cost-Optimized Architecture Design domain (15% of the exam guide), then 1 free question to test it. Part of the SAP-C03 study guide.
SAP-C03 describes a database’s traffic (steady, spiky, idle overnight, unpredictable after a launch) and its bill, and asks for the configuration and purchase option that costs least without throttling.
Amazon DynamoDB
Mode
Fits
Notes
On-demand
New, unpredictable or spiky traffic; tables idle much of the time
Pay per request; no capacity planning; optional maximum throughput setting caps cost
Provisioned with auto scaling
Predictable traffic that ramps gradually
Cheaper per request at steady utilization; can throttle on sudden spikes
Provisioned + reserved capacity
Steady, long-term provisioned throughput
1- or 3-year commitment for a set amount of capacity in one Region
Table classes:Standard-Infrequent Access lowers storage price but raises read and write price; pick it when storage, not throughput, dominates the table’s cost (old orders, history).
You can switch capacity mode, but only a limited number of times in a period, so it is not a minute-by-minute lever.
TTL deletes expired items at no write cost; DAX can cut read capacity for read-heavy, repetitive access.
Amazon RDS and Aurora
RDS Reserved Instances (1 or 3 years) discount steady instances. For most engines they are size-flexible within an instance family; they don’t apply across engines.
Aurora Serverless v2 scales capacity in fine-grained ACU steps and can scale to zero with auto-pause, which suits variable or intermittent workloads, dev and test, and multi-tenant databases with uneven load. A steadily busy database is often cheaper on provisioned instances with reserved pricing.
Aurora I/O-Optimized removes per-request I/O charges for a higher instance and storage price. AWS guidance is to consider it when I/O charges exceed about 25% of the cluster’s Aurora spend. Aurora Standard is cheaper for low or moderate I/O.
Graviton-based instance classes give better price performance for RDS and Aurora.
Stop idle dev databases, and use read replicas only where they serve real read traffic.
Database Savings Plans
A 1-year hourly spend commitment that applies across Aurora, RDS, DynamoDB, ElastiCache, DocumentDB, Neptune, Keyspaces, Timestream and DMS.
The discount follows usage across engines, Regions and deployment types, including moving from provisioned to serverless. AWS lists savings up to about 35% for serverless and about 20% for provisioned instances, with smaller savings on DynamoDB and Keyspaces throughput.
Choose it over engine-specific reserved instances when the company is modernizing or consolidating databases during the term; choose RIs or reserved capacity when one engine and instance will stay put and the RI discount is deeper.
How to choose
Scenario wording
Pick
“New app, traffic unknown, avoid throttling”
DynamoDB on-demand
“Steady predictable traffic for years”
DynamoDB provisioned with reserved capacity, or RDS RIs
“I/O charges are half the Aurora bill”
Aurora I/O-Optimized
“Database busy two hours a day”
Aurora Serverless v2
“Migrating from RDS to Aurora Serverless and DynamoDB this year, want a discount now”
Database Savings Plans
“Huge table of rarely read history”
DynamoDB Standard-IA table class
Common traps
Provisioned capacity sized for the peak on a spiky table.
I/O-Optimized for a low-I/O cluster.
Aurora Serverless v2 for a database at high, steady load all day.
Buying engine-specific RIs right before a planned engine migration.
Assuming Compute Savings Plans cover databases; they don’t.
A retailer runs an Amazon Aurora PostgreSQL cluster for its orders. The writer instance and two readers run around the clock, and the team expects that load to continue for at least three years.
A separate Amazon DynamoDB table holds shopping carts, and its traffic is very uneven, with sharp spikes during sales. The development and test databases are needed only during working hours.
Which TWO changes will reduce cost without reducing the capacity available to production? (Choose two.)
Choose two answers.
AA. Move the always-on Aurora writer and readers to Aurora Serverless v2, so that capacity is billed by the second as load changes.
BB. Buy reserved capacity for the Aurora writer and readers for three years, so their steady-state instance hours are billed at the committed rate.
CC. Set the cart table to on-demand capacity mode, so each spike is billed for the requests it actually uses without capacity planning.
DD. Set the cart table to provisioned capacity at the highest level seen during the last sale, so no spike is ever throttled.
EE. Keep the development databases running around the clock on their current instances, so that developers never wait for a database to start each morning.
Show the answer and explanation
Answers: B, C
A. Wrong: The cluster's load is steady, so the per-second billing of Serverless v2 would be paid around the clock at a higher effective rate than a commitment. This option raises cost for this pattern.
B. Right: Correct. Reserved capacity for a steady, three-year workload gives a lower committed rate on the instance hours, and the cluster keeps its current size. The commitment matches the expected load.
C. Right: Correct. On-demand capacity mode bills per request and absorbs uneven spikes without provisioning, which suits a cart table with sharp sales peaks. The table keeps its full capacity during spikes.
D. Wrong: A fixed level sized for the peak pays for that capacity all year, which wastes money in quiet periods. The option raises cost on a table with uneven traffic.
E. Wrong: Running non-production databases all day and night multiplies their instance cost for no benefit. Stopping them outside working hours is the cheaper approach.
Explanation
Reserved capacity fits the three-year steady Aurora load, and on-demand capacity mode fits the spiky cart table without over-provisioning. Serverless v2 costs more for steady load, while fixed peak capacity and always-on test databases waste money.
Want more on Database capacity modes and reserved capacity? The 5 full-length practice sets have 5 more Database capacity modes and reserved capacity questions, 375 in all, none of them in the free exam. One payment of $14.99, no subscription. See the practice sets →
Study Database capacity modes and reserved capacity