Course Introduction
PCA Standard
Exam Watch
Flashcards
Architectural guardrails, services, and trade-offs
Question
Question Text
Click to reveal answer
Answer
Answer Text
Architecture Decision Diagrams
Click a title to show/hide reference diagrams. Click images to expand.
Storage & Data Decision
Compute Decision Tree
Network Decision Tree
Security Layers Reference
Traffic & Load Balancer Selection
| LB Type | Scope | Protocol | Best For |
|---|---|---|---|
| HTTP(S) External | Global | HTTP/S | Web apps/APIs, CDN, WAF (Armor) |
| TCP/UDP External | Regional | TCP/UDP | Legacy protocols, non‑HTTP traffic |
| Internal HTTP(S) | Regional | HTTP/S | North‑south inside VPC, PSC backends |
| Internal TCP/UDP | Regional | TCP/UDP | Service‑to‑service within VPC |
| Network LB | Regional | L3/L4 | Low latency, passthrough needs |
Tip
Use Premium Tier for global HTTP(S) anycast; Standard
for regional cost savings.
Security
Attach Cloud Armor policies; use WAF rules; prefer
private NEGs + PSC for internal.
Streaming vs Batch Data Processing
| Pattern | Latency | Engine | Use Cases |
|---|---|---|---|
| Streaming | Sub‑second to seconds | Dataflow (Beam), Pub/Sub, BigQuery streaming | Telemetry, fraud, real‑time analytics |
| Micro‑batch | Minutes | Dataflow, Cloud Run Jobs, Workflows | Near‑real time aggregations |
| Batch | Hours | Dataproc/Batch, Dataflow, BigQuery scheduled | ETL, reports, backfills |
Throughput
Autoscale workers; use Shuffle/Streaming Engine; tune
windowing & triggers.
Storage
Hot: BigQuery/Bigtable; Cold: GCS; Catalog & govern
with Dataplex.
Compute Runtime Chooser
| Runtime | Ops Model | Best For | Notes |
|---|---|---|---|
| Cloud Run | Fully managed | Stateless HTTP, event‑driven | Scale to zero, rapid deploy, per‑request IAM |
| Cloud Functions | Fully managed | Simple event handlers | Lightweight, single‑purpose functions |
| GKE | Managed control plane | Complex microservices | Pods, HPA/VPA, mesh, service mesh policies |
| Compute Engine | Self‑managed | Custom/legacy workloads | Full control; use MIGs, images, OS Config |
Transactional vs Analytics Storage
| Service | Model | Strength | Choose When |
|---|---|---|---|
| Cloud Spanner | Relational | Global consistency, HA | Strict consistency + massive scale |
| Cloud SQL | Relational | Simplicity | Common RDBMS apps |
| Bigtable | Wide‑column | Low‑latency | Time‑series, large sparse datasets |
| Firestore | Document | Developer agility | Mobile/web with offline |
| BigQuery | Columnar analytics | Serverless analytics | BI/Reporting/ML over large data |
AI Solution Selection
| Option | Focus | Use When | Notes |
|---|---|---|---|
| Gemini APIs | LLM tasks | Chat, summarize, code assist | Fast start; enterprise controls |
| Vertex AI | ML ops | Train/tune/deploy/monitor | Pipelines, Model Garden, evaluation |
| Agent Builder | Agents | Task‑oriented workflows | Tool use, grounding, orchestration |
| Model Garden | Prebuilt | Pick best‑fit models | Integrate managed models |
Security
Use VPC SC, private endpoints, and data governance in
Dataplex.
Performance
Scale serving on GPUs/TPUs with Autoscaling; cache
prompts/responses when feasible.
Resilience & DR Topologies
| Topology | RTO/RPO | Complexity | Notes |
|---|---|---|---|
| Active‑Passive | Low/Low | Medium | Replicate data; DNS/LB failover |
| Active‑Active | Very low/Very low | High | Global LB; data sync; conflict resolution |
| Regional HA | Low/Low | Low | Zonal redundancy; snapshots/backups |
Hybrid Connectivity Choices
| Method | Availability | Bandwidth | Use When |
|---|---|---|---|
| Dedicated Interconnect | Very high | 10/100 Gbps | Mission‑critical hybrid, low latency |
| Partner Interconnect | High | 50 Mbps‑50 Gbps | Carrier‑mediated connectivity |
| HA VPN | High | Up to ~3 Gbps per tunnel | Quick setup, encrypted tunnels |
| Cloud VPN | Medium | Lower | Non‑HA or dev/test |
| VPC Peering | High | N/A | Privately connect VPCs (no transitive) |