Pair your devices with a code and playback position follows you: pause on this device, hit resume on the other. Position is saved to the site every minute and on pause.
Open this panel on your other device and enter the same code.
20 questions, four options each, one correct answer, no partial credit.
⚠️ Exam-styled questions written from the task statements and AWS documentation. Not the AWS Official Practice Question Set and not calibrated against real exam items.
1. A self-managed database on EC2 needs 150,000 IOPS with consistent sub-millisecond latency. Which storage meets the requirement?
2. A 100 GiB gp2 volume performs well for the first half hour of a batch job, then its IOPS drop sharply for the rest of the run. What is the most cost-effective fix?
3. Several Windows Server instances joined to Active Directory need a shared file system accessed over SMB. Which service fits?
4. An application writes all objects under a single S3 prefix and begins receiving 503 Slow Down errors at about 6,000 GET requests per second. What should the architect change?
5. An Auto Scaling group of workers processes messages from an SQS queue. Scaling on the number of visible messages causes over- and under-provisioning. What does AWS recommend?
6. A Lambda function that resizes images is CPU-bound and runs close to its timeout. How should the architect give it more CPU?
7. A research team runs tens of thousands of containerised simulation jobs with dependencies between them. Cost must be minimised and time-sensitive jobs must run first. Which service fits best?
8. A reporting workload slows down an RDS for MySQL production database. Reports can tolerate slightly stale data. What is the most appropriate solution?
9. A DynamoDB table must support 60 strongly consistent reads per second of 6 KB items. How many read capacity units are required?
10. A trading application on DynamoDB is write-intensive and every read must reflect the latest write. The team proposes DAX to reduce latency. What should the architect say?
11. An application needs an in-memory cache for a gaming leaderboard ranked by score, with automatic failover. Which is the best fit?
12. A company needs its partners to allowlist exactly two IP addresses for an application that runs behind Application Load Balancers in two Regions, with fast failover. Which service meets this?
13. A company orders a new dedicated 10 Gbps Direct Connect connection today, but must have encrypted connectivity from its data centre to its VPC within three days. What should it do?
14. A tightly coupled HPC application needs the lowest possible latency between nodes. Which combination is most appropriate?
15. An application subnet uses a /28 CIDR block. How many IPv4 addresses are available for
resources?
16. Several independent applications must each process the same clickstream events in real time, and must be able to replay the last three days of data. Which service fits?
17. IoT devices send CSV records. The data must land in S3 as Parquet with minimal code and no servers to manage. What is required?
18. On-premises applications must keep writing files over NFS, while the data is stored as objects in S3 and recently used files stay local for low latency. Which service fits?
19. Analysts query CSV files in S3 with Athena. Queries are slow and expensive, and most filter on a date column and read a few columns. Which change improves both?
20. A data lake must let analysts query a customer table in Athena without seeing the ssn column,
while the EU team sees only EU rows. Which service provides this?
1 — B. io2 Block Express goes to "256,000" IOPS with "average latency under 500 microseconds". gp3 (A) tops out at 80,000 and io1 (C) at 64,000. The 256,000 figure needs Nitro: "Other instance types … can achieve up to 32,000 IOPS." st1 (D) is HDD, optimised for throughput, not IOPS.
2 — B. Classic burst-credit exhaustion: baseline is 3 × 100 = 300 IOPS, burst is 3,000 until credits run out. gp3 gives "a consistent baseline IOPS performance of 3,000 IOPS" without bursting, lets you provision more independently of size, and costs "20 percent lower price per GiB". A buys ~6 TiB you don't need; C overshoots on cost; D is unrelated (and gp2 doesn't support Multi-Attach).
3 — B. "Using Amazon EFS with Microsoft Windows–based Amazon EC2 instances is not supported." FSx for Windows uses "Server Message Block (SMB)" and "Users accessing file systems are authenticated with Microsoft Active Directory." C is Linux/HPC; D is block storage, not a file share.
4 — B. "at least 3,500 PUT/COPY/POST/DELETE or 5,500 GET/HEAD requests per second per partitioned Amazon S3 prefix. There are no limits to the number of prefixes." Spreading across prefixes raises the aggregate; 503s appear while scaling. A addresses distance, not request rate. C changes AZ resilience and Region for no reason.
5 — B. AWS: "the number of messages in your SQS queue does not solely define the number of instances needed"; use "a backlog per instance metric with the target value being the acceptable backlog per instance", computed as acceptable latency ÷ average processing time, with target tracking.
6 — B. "Lambda allocates CPU power in proportion to the amount of memory configured." There is no vCPU parameter (A). AWS calls memory "the principal lever for controlling the performance of a function"; neither C nor D changes the CPU allocated to an invocation.
7 — B. Batch "automatically provisions compute resources"; job queues carry priority ("a high priority queue that you submit time-sensitive jobs to"); jobs "can be dependent on the successful completion of other jobs"; managed compute environments take "the Spot Instance price threshold". Lambda (C) is capped at 15 minutes and has no job dependency model; D is a long-running service shape.
8 — B. Read replicas let you "scale out beyond the capacity constraints of a single DB instance for
read-heavy database workloads", and AWS lists "Business reporting or data warehousing scenarios".
Replication is asynchronous, which the stem tolerates. A fails: a Multi-AZ standby cannot serve reads
(SAA2). C fixes connection counts, not query load. D may help but doesn't separate the workloads.
9 — C. 6 KB ÷ 4 KB = 1.5 → round up to 2 RCU per strongly consistent read. 2 × 60 = 90. Eventually consistent would be 45 — not an option, and not what was asked.
10 — B. DAX is not ideal for "Applications that require strongly consistent reads" or "Applications that are write-intensive". It serves "eventually consistent" data. C and D are irrelevant; DAX performs best above 90% hit rate.
11 — B. Leaderboards need sorted sets; failover needs replication. The comparison table shows Memcached with no sorted sets, no replication and no automatic failover; Valkey/Redis OSS have all three. DAX (C) is DynamoDB-only; EFS (D) isn't a cache.
12 — B. Global Accelerator provides "two static IPv4 addresses" anycast from the edge, fronting ALBs in multiple Regions, and "reacts instantly to changes in health". CloudFront (A) gives a domain name, not fixed IPs. D gives one IP per subnet per Region — more than two addresses, and no cross-Region failover.
13 — B. A dedicated connection can take "up to 72 business hours for AWS to review your request and provision a port", before the physical cross-connect. Site-to-Site VPN is IPsec and quick to create. C is wrong twice: hosted connections are provisioned by a partner, and "You can't request a hosted connection through the Direct Connect console." D is not a data-centre option.
14 — B. Cluster placement groups give "low-latency network performance necessary for tightly coupled node-to-node communication", inside one AZ. EFA provides "OS-bypass" for MPI. A and C are for reducing correlated failures, not latency. D adds hops.
15 — C. A /28 has 16 addresses; "The first four IP addresses and the last IP address in each
subnet CIDR block are not available for your use" → 16 − 5 = 11.
16 — B. "There can be multiple applications for one stream, and each application can consume data from the stream independently and concurrently." Retention defaults to 24 hours and can be raised to 8,760. SQS (A) gives each message to one consumer. Firehose (C) delivers; it isn't read by independent consumers with replay.
17 — B. Firehose converts "from JSON to Apache Parquet or Apache ORC" using a schema from the Glue Data Catalog. "If you want to convert an input format other than JSON, such as comma-separated values (CSV) … you can use AWS Lambda to transform it to JSON first." A misses the CSV step.
18 — B. S3 File Gateway stores files as S3 objects over "NFS version 3 or 4.1" and "provides low-latency access to data through transparent local caching". DataSync (A) copies; it doesn't present a live mount. C is iSCSI block storage. D is for partner file transfer protocols.
19 — B. Partitioning means "only data from matching partitions is read"; columnar formats mean "Only the columns needed for the query are loaded". CTAS or a Glue job converts in one step. C is the opposite of AWS's advice — "Datasets that consist of many small files result in poor overall query performance." A and D don't apply to Athena.
20 — B. Lake Formation permissions "are enforced using granular controls at the column, row, and cell-levels across … Amazon Athena", with data filters for "row and cell-level security". Bucket policies (A) govern objects, not table columns. Macie (C) discovers sensitive data; it doesn't enforce query-time column access.
| Score | Read this as |
|---|---|
| 18–20 | Domain 3 is solid. Move on; revisit a week before the exam. |
| 14–17 | Good. Re-read the lesson behind each miss — the explanations name them. |
| 10–13 | You know the services but not the limits. Redo the cheat sheet, then lessons 1 and 3. |
| < 10 | Work the module in order and do the lab before re-testing. |
Miss patterns: