AWS · Domain
Core architecture
Storage, databases, scaling out, containers, decoupling, and server migration: the 7 Rs, Application Migration Service, Application Discovery Service and Migration Hub. 6 worlds, 36 stages, about 16 hours to complete.
Architecture World 1 · Associate
Block, File, Object
The storage type comes first and the limits are arithmetic: IOPS per GiB, requests per prefix, days at line rate, hours since the last backup.
- 1Volumes That FitFreeFour volumes for the rehosted booking database, one instance to carry them. Make the meter meet every workload at the lowest bill.
- 2Shared by ManyFive populations want a shared file system. Each has one requirement that picks its service.
- 3Objects at ScaleTwelve ships upload CCTV to one prefix at once, a partner downloads timetables on Harbourline's bill, and Singapore is always late. Run the objects through S3's numbers.
- 4Bridge to the Harbour400 TB must leave Portsmouth in 30 days over a 500 Mbps link, and three applications are staying behind. Do the days-at-line-rate arithmetic, then pick the bridge.
- 5Back It UpEvery resource has an RPO and an RTO on paper and a nightly snapshot in practice. Make the interval match the promise, and keep a copy where the production account cannot reach.
- 6Storage DeskFive storage tickets from Monday morning. Read the metric before you touch a setting.
Architecture World 2 · Associate
Right Database
The access pattern picks the engine, the failure picks the deployment, the arithmetic picks the capacity, and the cache answers what the replica cannot.
- 1Pick the EngineEight workloads, one database each. The access pattern decides; the licence and the migration tool follow.
- 2Survive the ZoneAvailability Zone b goes dark at 14:02. Watch which deployment keeps writing, which keeps reading, and how long the application waits.
- 3Aurora's WayAurora answers availability with a shared volume and up to fifteen replicas. Set the tiers, the endpoints and the Regions, then watch the writer fall over.
- 4Capacity UnitsThe timetable throttles on one key while the table has capacity to spare. Do the arithmetic the exam asks: units, items, partitions, indexes.
- 5Cache ItSix read problems, six answers. A cache, a durable in-memory store, a proxy or the writer: the symptom decides.
- 6Database DeskSix database tickets. The metric names the cause before the fix does.
Architecture World 3 · Professional
Elastic Fleet
Capacity follows a signal with a delay, a load balancer only sends traffic where a health check says, and resilience is capacity you already paid for in the zone or Region you have not lost yet.
- 1Pick the ComputeNine workloads, each with a shape. Pick the compute, the purchase option and the placement, and watch the bill under the board.
- 2Balance the LoadSeven sample requests through two load balancers. Each one lands on a rule, a target group, an Availability Zone and a target, or on the reason it did not.
- 3Scale on a SignalA day of bookings on the clock. Set the policy, the warmup and the grace period, and watch the red area between demand and capacity shrink.
- 4Lose a ZoneAvailability Zone b goes dark at 09:14. Which fleet keeps serving, whose sessions survive, what comes back by itself, and how fast does traffic leave the dark zone?
- 5Across RegionsFour workloads, four rungs of the DR ladder, then the recovery itself: order the steps and watch the RTO add up.
- 6Fleet DeskEight fleet tickets. Target health, activity history and the metrics say why before you change anything.
Architecture World 4 · Professional
Shipped in Containers
Two layers scale and two roles sign: the task and the host, the role that pulls the image and the role the application calls with.
- 1Pick the PlatformSeven workloads want containers, or think they do. Pick who runs the hosts, who runs the scheduler, and which one should not be a container at all.
- 2Task and ServiceOne task definition, two roles, one network interface per task. Make the manifest service pull, log, call S3 and answer the load balancer, then let a bad deploy roll itself back.
- 3Pods and NodesThe pricing engine's cluster: who gives the pods their permissions, their volumes, their addresses and their nodes.
- 4Images and RegistriesImages through the registry's rules: which stay, which are refused, which are scanned, which reach Ireland, and which a private subnet can pull at all.
- 5Containers That ScaleTwo layers on the clock: tasks scale on requests, hosts scale under them, and the second delay is the one that hurts.
- 6Container DeskSeven container tickets. The stopped reason, the service event and the pod's events name the cause.
Architecture World 5 · Associate
Loose Coupling
A queue turns a burst into a backlog, a topic turns one event into many, and every serverless limit is a number the clock can reach.
- 1Queue or TopicEight integrations, one service each. Order, exactly once, fan-out, filtering, replay, protocol, duration: the requirement names the service.
- 2Scale on the BacklogA cancellation storm: 400 rebookings in a minute. Scale the workers on the backlog, and set the queue's numbers so nothing is processed twice, lost or stuck.
- 3Fan OutBooking events as tokens through a topic, a bus and a pipe. Each one is delivered, filtered, dropped or replayed by a rule you can read.
- 4Functions Under LoadInvocations per second against Lambda's numbers. One pool for every function, a rate for every burst, and a duration that memory can buy down.
- 5Front DoorSix calls through API Gateway. The API type, the endpoint type, the throttle, the timeout and the cache decide each one.
- 6Decoupling DeskSeven integration tickets. The queue's age, the function's throttles, the rule's matches and the API's codes name the cause.
Architecture World 6 · Professional
Move and Modernise
Every server gets one of seven verbs, replication runs before the cutover, and the bill after the move is a number the player can lower.
- 1The Seven RsEight applications, seven verbs. The facts the exam gives decide which: age, licence, users, coupling, deadline, value.
- 2Discover and TrackThirty-eight servers and a 40-VM cluster, three ways to see them, one place to track them. Pick the tool that sees what each estate allows.
- 3Lift and ShiftEight booking servers, one cutover weekend. Order the steps so replication is steady before the launch, the test comes before the cutover, and the source stays until it is final.
- 4Pay Less, Size RightThe first month's bill after the move, as a meter. Rightsize, commit, alert, allocate, and watch the number fall.
- 5Modernise ItThe rehosted booking monolith, modernised a slice at a time. Put a door in front, move one path, and keep everything else serving.
- 6Migration DeskSeven migration tickets. Replication status, flow logs and the first bill name the cause.