Memory hook: Elasticity follows demand; availability survives a failure; agility shortens the time to try an idea.
Must remember
- Cloud computing supplies technology on demand with usage-based charging. Replace large upfront capacity purchases with variable spending; economies of scale can reduce unit cost. Neither eliminates waste automatically.
- Scalability handles growth by increasing capacity. Vertical scaling enlarges a machine; horizontal scaling adds machines. Elasticity adds and removes capacity as demand changes. A load balancer distributes requests; it does not create capacity.
- A Region is a geographic deployment area. An Availability Zone is an isolated failure domain containing one or more data centres. Multiple AZs support regional availability; multiple Regions address regional disasters or geographic requirements. An edge location brings eligible content or traffic processing closer to users.
- Choose a Region using residency requirements, latency, feature availability and price. A nearby Region without the required service is not automatically suitable. Global services still have data and resource-location rules.
- IaaS gives more infrastructure control, PaaS removes more platform management, SaaS delivers a finished application. Public cloud, private/on-premises and hybrid describe deployment arrangements, not permission levels on an S3 bucket.
- Memorise the six Well-Architected pillars through questions: can we operate it well; protect it; recover it; perform efficiently; avoid waste; and reduce environmental impact? These are operational excellence, security, reliability, performance efficiency, cost optimisation and sustainability.
- Design for failure, automate repeatable work, measure before changing capacity and use managed services when the required control permits. High durability protects stored data; high availability keeps a service usable.
Choose under exam pressure
| Requirement | Choice and reason |
|---|---|
| Survive loss of a data centre | Deploy useful capacity across AZs. |
| Worldwide static content | CloudFront edge caching. |
| Rapid experiments without buying servers | Cloud agility and variable spending. |
Traps
- An AZ is not a synonym for a single building.
- Multi-AZ does not itself protect against loss of the entire Region.
- Pay as you go is a billing model, not a promise that every resource stops charging when unused.
Active recall
1. Traffic rises every evening and falls at midnight. Which property matters?
Elasticity: capacity should increase and decrease with demand.
2. A compliance rule restricts data to one country. What comes first?
Eligible data locations and service features; optimise latency and cost within that constraint.
3. Does a second server in the same AZ solve AZ failure?
No. Place sufficient application capacity in another AZ and address state and routing too.
4. Which pillar asks whether an oversized idle server is wasteful?
Cost optimisation; reducing unnecessary resources also supports sustainability.
5. Is a backup equivalent to an available application?
No. A backup preserves recoverable data; availability also needs functioning compute, dependencies and access.