5. Service Selection Guide

Data services decision table

AI services decision table

Compute and modernization decision table

Cost and operations decision table

6. Architecture Patterns

Pattern 1: Streaming analytics and dashboards

Event producers
      ↓
   Pub/Sub
      ↓
   Dataflow
      ↓
   BigQuery
      ↓
    Looker

Use when: Events arrive continuously and leaders need timely dashboards or insights.

Why it works:

• Pub/Sub ingests messages asynchronously.
• Dataflow processes streaming data.
• BigQuery stores and analyzes data.
• Looker exposes governed BI.

Trap: Do not assign the ingestion job to Looker or the dashboarding job to Pub/Sub.

Pattern 2: Global transactional application

Global application
      ↓
 Cloud Spanner

Use when: The application needs relational transactions, strong consistency, and global horizontal scalability.

Trap: Cloud SQL is relational, but the stated global scaling requirement points to Cloud Spanner.

Pattern 3: Mobile or web document application

Mobile or web app
      ↓
   Firestore

Use when: The app needs a flexible document model and application-friendly access patterns.

Trap: Do not select Firestore merely because the application is modern. Select it because the document model fits.

Pattern 4: Object storage lifecycle

Frequently accessed objects → Standard
Monthly retrieval          → Nearline
Quarterly retrieval        → Coldline
Yearly or rarer retrieval  → Archive

Use when: The question focuses on access frequency and storage cost.

Pattern 5: Serverless event handling

Cloud Storage upload
      ↓
Cloud Functions
      ↓
Business action

Use when: Code should react to an event without provisioning servers.

Pattern 6: Serverless container application

Containerized HTTP application
      ↓
    Cloud Run
      ↓
Automatic scaling with reduced infrastructure administration

Use when: The application is packaged as a container, and Kubernetes-level control is not required.

Pattern 7: Container orchestration

Complex containerized applications
      ↓
      GKE
      ↓
Managed Kubernetes orchestration

Use when: Teams explicitly need Kubernetes or complex container orchestration.

Pattern 8: API ecosystem

Internal business capabilities
      ↓
Apigee API Management
      ↓
Partners, developers, and controlled external consumers

Use when: APIs need publishing, security, analytics, or monetization.

Pattern 9: Hybrid environment

On-premises systems + public cloud services
      ↓
Hybrid-cloud strategy

Use when: Some systems remain outside the public cloud due to regulation, compatibility, or staged modernization.

Pattern 10: Multicloud management

Multiple cloud environments + hybrid infrastructure
      ↓
    Anthos
      ↓
Unified management approach

Use when: The exam scenario explicitly asks for centralized hybrid or multicloud infrastructure management.

7. Exam Traps

Trap 1: Choosing the most advanced service

A more powerful service is not automatically the right service.

• Choose Cloud SQL for a conventional managed relational database.
• Choose Cloud Spanner when global scalability and strong consistency are explicit requirements.
• Choose Cloud Run when serverless containers meet the need.
• Choose GKE when Kubernetes orchestration is truly required.

Trap 2: Confusing an analytics warehouse with an operational database

• BigQuery: analytics warehouse
• Cloud SQL: conventional relational application database
• Cloud Spanner: globally scalable relational transactions
• Cloud Bigtable: low-latency wide-column workloads
• Firestore: document-oriented application data

Trap 3: Confusing messaging, processing, analytics, and visualization

• Pub/Sub transports events.
• Dataflow processes data.
• BigQuery analyzes data.
• Looker presents governed insights.

Trap 4: Defaulting to custom AI

For a standard task:

1. Start with a pre-trained API.
2. Use AutoML when your own data should improve or tailor the model.
3. Use a custom Vertex AI model when differentiation justifies added effort and expertise.

Trap 5: Confusing speech directions

• Audio → text = Speech-to-Text
• Text → audio = Text-to-Speech

Trap 6: Confusing hybrid and multicloud

• Hybrid: on-premises or private environment plus public cloud
• Multicloud: more than one cloud provider

Trap 7: Confusing scalability and elasticity

• Scalability: handle increased demand
• Elasticity: adjust capacity as demand changes

Trap 8: Assuming cloud eliminates customer security responsibilities

The shared responsibility model changes the division of work. It does not eliminate customer responsibilities.

Trap 9: Confusing authentication, authorization, and auditing

• Authentication: verify identity
• Authorization: determine permissions
• Auditing: record activity

Trap 10: Confusing residency and sovereignty

• Residency: geographic storage location
• Sovereignty: jurisdictional authority and applicable law

Trap 11: Confusing Cloud Armor and IAM

• Cloud Armor protects applications from network and web attacks.
• IAM controls identities and resource permissions.

Trap 12: Confusing budgets and quotas

• Budget threshold rules notify stakeholders about spending.
• Resource quotas restrict consumption.

Trap 13: Treating high availability and disaster recovery as synonyms

• High availability minimizes disruption.
• Disaster recovery restores service after major incidents.

Trap 14: Migrating every workload the same way

A portfolio can contain applications to:

• Retire
• Retain
• Rehost
• Replatform
• Refactor
• Reimagine

Trap 15: Forgetting the business requirement

When several answers are technically possible, choose the one that most directly satisfies the stated business need with appropriate complexity.

Option-elimination technique

Eliminate answers in this order:

1. Wrong category: A security tool cannot replace a database.
2. Wrong workload: A BI tool is not an ingestion service.
3. Too broad or too complex: GKE may be unnecessary when Cloud Run meets the need.
4. Incomplete fit: Cloud SQL is relational, but it is not the best answer for global horizontal scaling with strong consistency.
5. Absolute statement: Options claiming that cloud “eliminates all responsibilities” or “guarantees zero downtime” are usually wrong.

8. Quick Memory Rules

Cloud fundamentals

• Elasticity expands and contracts.
• Scalability supports growth.
• Agility speeds experimentation and delivery.
• CapEx buys infrastructure. OpEx consumes services.
• TCO means lifecycle cost, not list price.
• Hybrid = home plus cloud. Multicloud = multiple providers.
• DNS resolves names. Latency is delay. Bandwidth is volume over time.

Data

• Cloud Storage stores objects.
• Cloud SQL = conventional relational database.
• Cloud Spanner = globally scalable relational transactions.
• Cloud Bigtable = massive low-latency wide-column data.
• Firestore = flexible document database.
• BigQuery = serverless analytics warehouse.
• Looker = governed dashboards and BI.
• Pub/Sub receives events. Dataflow processes streams.
• Standard active; Nearline monthly; Coldline quarterly; Archive yearly or rarer.

AI

• Common task, fast start: pre-trained API.
• Your data, reduced specialist effort: AutoML.
• Differentiated proprietary requirement: custom Vertex AI model.
• SQL analysts building ML in the warehouse: BigQuery ML.
• Images: Vision. Text insight: Natural Language. Languages: Translation.
• Speech-to-Text listens. Text-to-Speech speaks.

Modernization

• Rehost relocates.
• Replatform improves.
• Refactor redesigns.
• Reimagine reinvents.
• Compute Engine = VMs.
• Cloud Run = serverless containers.
• Cloud Functions = event-driven code.
• App Engine = managed application platform.
• GKE = managed Kubernetes.
• Apigee = API management.
• Anthos = hybrid and multicloud management.

Security and operations

• Authenticate identity. Authorize actions. Audit evidence.
• IAM controls access. Cloud Armor protects web-facing applications.
• Residency is location. Sovereignty is jurisdiction.
• Quota limits usage. Budget alerts on spend. Billing Reports explain spend.
• Fault tolerance continues. High availability minimizes downtime. Disaster recovery restores.
• DevOps collaborates and automates. SRE engineers reliability.

9. Final Revision Notes

The 30 facts to know cold

1. Cloud adoption is a business transformation decision, not merely server relocation.
2. Elasticity means resources expand or contract as demand changes.
3. Scalability means a system can handle more load.
4. Cloud consumption commonly shifts spending from large CapEx purchases toward OpEx.
5. TCO includes staffing, maintenance, infrastructure, and lifecycle operations.
6. Hybrid cloud combines on-premises or private systems with public cloud.
7. Multicloud uses more than one cloud provider.
8. IaaS provides infrastructure control; PaaS abstracts more platform operations; SaaS provides a finished application.
9. Shared responsibility never means the customer has no security duties.
10. Cloud Storage is object storage.
11. Cloud SQL is a conventional managed relational database.
12. Cloud Spanner is for globally scalable relational transactions with strong consistency.
13. Cloud Bigtable is for large, low-latency wide-column workloads.
14. Firestore is a NoSQL document database for application experiences.
15. BigQuery is a serverless managed analytics warehouse.
16. Looker supports governed self-service BI.
17. Pub/Sub ingests events; Dataflow processes them.
18. Good AI outcomes require good data and responsible AI practices.
19. Use a pre-trained API before building custom AI unless differentiation is required.
20. BigQuery ML enables SQL-based ML in BigQuery.
21. Rehost is lift and shift; replatform is move and improve; refactor is redesign.
22. Compute Engine provides VMs.
23. Cloud Run is serverless containers; Cloud Functions is event-driven code; GKE is managed Kubernetes.
24. APIs can create partner ecosystems and new revenue opportunities.
25. Apigee manages APIs; Anthos manages hybrid and multicloud infrastructure.
26. IAM controls access; Cloud Armor helps protect against DDoS and web attacks.
27. Authentication, authorization, and auditing are different.
28. Budget thresholds alert; quotas limit resource consumption.
29. High availability, fault tolerance, and disaster recovery are related but distinct.
30. Efficient cloud use and informed architecture decisions can support sustainability goals.

Last-minute comparison grid

Five-question self-test

1. A company needs a managed SQL database for a conventional regional application.
   Answer: Cloud SQL.

2. A company wants to transcribe contact-center recordings.
   Answer: Speech-to-Text API.

3. A legacy application must move quickly with almost no redesign.
   Answer: Rehost.

4. A public web application needs protection from DDoS attacks.
   Answer: Cloud Armor.

5. Finance wants notifications before cloud spend exceeds expectations.
   Answer: Budget threshold rules.

10. Exam-Day Checklist

Before the exam

• Review the official exam guide one final time for current product names.
• Confirm the test time, delivery method, identity requirements, and technical setup.
• Sleep adequately and avoid last-minute deep study.
• Keep the final comparison grid fresh in memory.

During the exam

• Read the last sentence of the question first to identify the requested decision.
• Underline mentally the primary constraint: global, serverless, relational, real time, monthly access, least privilege, DDoS, hybrid, or minimal operations.
• Eliminate answers from the wrong category.
• Prefer the simplest service that satisfies the requirement.
• Watch for absolute claims such as “always,” “every,” “guarantees,” or “eliminates all responsibility.”
• Distinguish a related concept from the best answer.
• Mark uncertain questions and return after completing the easier ones.
• Do not over-engineer business scenarios.

Final five-minute review

Confirm that you have not confused:

• Hybrid cloud and multicloud
• Scalability and elasticity
• BigQuery and Cloud SQL
• Cloud SQL and Cloud Spanner
• Pub/Sub and Dataflow
• Speech-to-Text and Text-to-Speech
• Cloud Run and GKE
• IAM and Cloud Armor
• Authentication and authorization
• Budgets and quotas
• High availability and disaster recovery

End of course