0% completed
The Well-Architected Framework and Cloud Service Mapping
On This Page
The Six Pillars
How an SA Runs a Review
The Retailer's Review After the Second Wave
Frameworks From Other Providers
Cloud Service Mapping
Using This in an Interview
Key Takeaways
Practice Questions
The AWS Well-Architected Framework is a published set of questions and best practices that a team uses to review one workload against six areas of quality, called pillars. A review with it does not grade the design. It produces a list of risks, each with an owner and a planned fix.
The framework matters to an SA for two reasons. It gives the review after each migration wave a fixed structure, so nothing important is skipped. It also gives the SA and the customer's engineers a shared vocabulary for trade-offs, such as reliability against cost.
This lesson covers the six pillars, how an SA runs a review, how the other large clouds compare, and how common services map across AWS, Azure, and Google Cloud.
The Six Pillars
Each pillar asks whether the workload meets one kind of need, and a good design balances all six. The questions in the table are written in the style of the framework rather than quoted from it, and the findings are common ones that SAs report.
| Pillar | A typical review question | A common finding |
|---|---|---|
| Operational excellence | How do you know the workload is healthy, and how do you deploy changes safely? | Changes are deployed by hand, and no one can roll one back quickly |
| Security | How do you manage identities and permissions for people and programs? | Long-lived access keys with administrator rights |
| Reliability | How do you back up data, and have you tested a restore? | Backups exist but a restore has never been tried |
| Performance efficiency | How did you choose the server types and database for this workload? | Server types copied from the data center and never tested under load |
| Cost optimization | Who watches spending, and how is it assigned to teams? | No cost labels, so no team knows what it spends |
| Sustainability | How do you limit the resources the workload consumes? | Test environments run every night and weekend with no one using them |
The pillars often conflict. A standby copy in a second region raises reliability but also raises cost. The framework does not choose for the customer; it makes the trade-off visible so that the customer can choose.
How an SA Runs a Review
A review covers one workload at a time, with the people who build and run it, and ends with an improvement plan. The SA usually follows five steps:
- Choose the workload and its owner. A review of "everything" produces general advice that no team acts on.
- Collect facts before the meeting. These include the architecture diagram, the RPO and RTO, the last incidents, and the monthly bill.
- Go through the questions with the engineers. The SA asks how each practice is done today, not whether the team agrees with it.
- Rate each risk as high or medium. AWS's Well-Architected Tool, a free service in the AWS console, records answers and labels risks this way.
- Agree on the fixes with owners and dates, then review the same workload again after the fixes or after a large change.
A review is a conversation with the customer's engineers, not an audit. When engineers feel judged, they describe the design they planned rather than the one that runs.
The Retailer's Review After the Second Wave
After the second wave moved the store and warehouse systems, the SA reviewed them with the retailer's engineers. The reliability questions found that two applications had no tested backup. One had no backup schedule at all, and the other copied its data every night but had never restored it.
The SA rated both as high risk. The team then set an RPO of 24 hours and an RTO of 8 hours for both applications, scheduled daily backups to a second region, and ran a restore test. The first restore took 5 hours, inside the 8-hour target, and the team wrote the steps down so that any engineer could repeat them.
The same review found a medium risk under cost optimization: test servers for the warehouse system ran all weekend. A schedule that stops them outside working hours was added to the improvement plan.
Frameworks From Other Providers
Azure and Google Cloud publish their own frameworks with similar content. The Azure Well-Architected Framework has five pillars: reliability, security, cost optimization, operational excellence, and performance efficiency. Google Cloud publishes the Google Cloud Well-Architected Framework, which also groups its guidance into pillars.
An SA who knows one framework can work with the others, because the questions cover the same concerns. Use the framework of the customer's main platform, so that its guidance and tools match the services in the design.
Cloud Service Mapping
Most common services have a close equivalent on each large cloud, but the details differ. The table maps twelve of them. Limits, pricing models, and features differ between the columns, so check the specific service before promising equal behavior.
| Need | AWS | Azure | Google Cloud |
|---|---|---|---|
| Virtual servers | EC2 | Azure Virtual Machines | Compute Engine |
| Managed Kubernetes, which runs containers (packaged applications) | EKS | Azure Kubernetes Service (AKS) | Google Kubernetes Engine (GKE) |
| Serverless functions, code run per event with no servers to manage | Lambda | Azure Functions | Cloud Run functions |
| Object storage, files reached over the web by name | S3 | Azure Blob Storage | Cloud Storage |
| Block storage, disks attached to servers | EBS | Azure Managed Disks | Persistent Disk |
| Managed relational database | RDS | Azure SQL Database, Azure Database for PostgreSQL | Cloud SQL |
| Key-value and document database, records fetched by key | DynamoDB | Azure Cosmos DB | Firestore |
| Data warehouse, a database for large reporting queries | Redshift | Azure Synapse Analytics | BigQuery |
| Private network, isolated from other customers | VPC | Azure Virtual Network | VPC |
| Dedicated link to a data center | Direct Connect | ExpressRoute | Cloud Interconnect |
| Content delivery network, copies of files near users | CloudFront | Azure Front Door | Cloud CDN |
| Key management service | KMS | Azure Key Vault | Cloud KMS |
The mapping helps most when a customer uses two clouds, or when a candidate has worked on a different cloud from the interviewer's. It helps least when the customer depends on a feature that only one service has.
Using This in an Interview
When an interviewer asks how you would review an existing system, describe the method and one real finding.
"I would review one workload at a time against the six Well-Architected pillars, with its engineers in the room. For a retailer, the reliability questions showed that two applications had no tested backup. We set an RPO and RTO, scheduled backups to a second region, and ran a restore that took 5 hours against an 8-hour target. Every finding had an owner and a date."
Expect two follow-up questions:
- "The customer uses Azure. Does your method change?" No. The questions cover the same concerns, so use the Azure framework and its service names.
- "What if the customer will not fix a high risk?" Write down the risk and its likely effect, and ask the customer's executive to accept it in writing.
Key Takeaways
- The Well-Architected Framework reviews one workload against six pillars: operational excellence, security, reliability, performance efficiency, cost optimization, and sustainability.
- A review produces rated risks with owners and dates, not a grade.
- The pillars often conflict, and the review makes those trade-offs visible.
- Azure and Google Cloud publish similar frameworks, and most common services have a close equivalent on each cloud.
The next lesson, What the Architecture Design Round Tests, shows how interviewers test these foundations when you design a system for a customer.
Practice Questions
1. A customer asks the SA to review "all our cloud systems" in one meeting. What should the SA propose?
<details><summary>Show answer</summary>A review of one important workload first, with its owner and engineers. A review of everything produces general advice that no team acts on. One workload gives specific risks, named owners, and fixes that the team can finish.
</details>2. Why did the retailer run a restore test instead of only checking that backups existed?
<details><summary>Show answer</summary>Because a backup that has never been restored is not proven to work. The test measured the real recovery time, 5 hours against an 8-hour target, and produced written steps that any engineer can follow.
</details>3. A customer on Google Cloud asks which service matches AWS Lambda. What do you answer, and what do you add?
<details><summary>Show answer</summary>Cloud Run functions is the closest match. Add that limits such as the maximum run time and the supported triggers differ, so the design should be checked against the specific service before the customer commits.
</details>Reading Progress
0%
On This Page
The Six Pillars
How an SA Runs a Review
The Retailer's Review After the Second Wave
Frameworks From Other Providers
Cloud Service Mapping
Using This in an Interview
Key Takeaways
Practice Questions