Networking and content delivery
A VPC is your own network in AWS. Everything else in this area is either a way in, a way out, a way to connect two networks, or a way to get closer to users.
Exam tip. Cacheable HTTP content means CloudFront. Non-HTTP traffic, static entry addresses or fast regional failover means Global Accelerator.
Choosing a compute option
The compute spectrum runs from virtual machines you administer to functions where only code is yours. Less to operate usually means less control and a narrower fit, so the question is which the workload actually needs.
Exam tip. Commit to the baseline and flex the peak. Committing at peak pays for idle capacity all year; committing before right-sizing locks in the waste.
Choosing a database
AWS offers a purpose-built database per data shape. Matching the access pattern to the engine is the whole exercise; the cost is more systems to operate, which is why a general-purpose relational database is still often right.
Exam tip. Replication is not backup. A global table or read replica copies a bad write everywhere, which is why point-in-time recovery exists.
Decoupling and serverless patterns
Loose coupling lets components fail and scale independently. The exam rewards designs where a producer does not wait on, or even know about, its consumers.
Exam tip. Ask what the user is actually waiting for. Anything not needed for their response belongs on the asynchronous side of a queue or event.
Choosing a storage service
Three shapes of storage answer three different access patterns. Object storage for whole items over HTTP, block storage for a disk, file storage for a shared mount. The access pattern decides, not the size of the data.
Exam tip. Cheaper per gigabyte comes with strings: minimum storage duration, minimum object size and retrieval charges. Short-lived data belongs in Standard.
Analytics, machine learning and AI services
Analytics splits into preparing data, querying it and showing it. AI services split by what you feed them. Matching the input type or the pipeline stage to the service answers most questions in this area.
Exam tip. Analytics on operational data should not run against the operational store. Export it, then query the copy.