Analyze and Optimize Technical and Business Processes
This domain accounts for 18% of the GCP Professional Cloud Architect exam and tests your ability to apply cost management, performance optimization, and business continuity strategies to real GCP architectures. Expect scenario-based questions where you must select the most cost-effective, performant, or resilient design from a set of plausible alternatives. The exam emphasizes knowing which GCP tool or pattern solves a specific operational problem, and why the alternatives are wrong.
Cloud Billing: Budgets, Alerts, and BigQuery Export
Core Concepts
Cloud Billing budgets set spending thresholds and trigger alerting notifications via email or Pub/Sub. They do NOT automatically stop spending—they only alert. Budget scope can be set at the billing account, project, product, label, or credit type level.
Default alert thresholds: 50%, 90%, and 100% of the budget amount. You can add up to five custom thresholds. Alerts can target billing admins/users (email) or a Pub/Sub topic for programmatic response (e.g., triggering a Cloud Function to disable a billing account via the Cloud Billing API).
Billing export to BigQuery enables detailed cost analysis. Two export types:
| Export Type | Granularity | Use Case |
|---|---|---|
| Standard usage cost | Daily | Cost breakdown by service, project, SKU |
| Detailed usage cost | Hourly, resource-level | Resource-level attribution, chargeback |
| Pricing export | Published list prices | Cost modeling, forecast tooling |
Data lands in a BigQuery dataset you specify. You can then use Looker Studio (formerly Data Studio) or custom SQL to build dashboards.
Common Traps
- Budgets do not cap spending automatically. To stop spending programmatically, you must wire a Pub/Sub alert to a Cloud Function that calls the Billing API.
- Billing export is not retroactive—data appears only after export is enabled.
- You cannot export billing data to Cloud Storage directly from the billing console; the native destination is BigQuery.
- Budget alerts fire based on actual or forecasted spend; you must choose which trigger type when creating the budget.
Worked Scenario
Requirement: A team wants to be notified when their project is 80% through a $1,000 monthly budget, and automatically disable billing if the budget is exceeded.
Reasoning: Create a budget with a custom 80% alert threshold sending to a Pub/Sub topic. Wire a Cloud Function subscribed to that topic. For the 100% threshold, a second Pub/Sub message triggers the Cloud Function to call billingAccounts.projects.updateBillingInfo with an empty billing account, effectively disabling billing for the project.
Cost Optimization: CUDs, SUDs, and Rightsizing
Committed Use Discounts (CUDs)
CUDs are 1-year or 3-year commitments to use a minimum level of Compute Engine resources in exchange for discounted prices.
| CUD Type | Discount | Flexibility | Applies To |
|---|---|---|---|
| Resource-based | Up to 57% (1-yr), 70% (3-yr) | Machine type-specific | Compute Engine VMs |
| Spend-based | Up to 25% (1-yr), 52% (3-yr) | Any usage of that service | Cloud Run, GKE, Cloud SQL |
CUDs are regional commitments. They can be shared across projects in a billing account via commitment sharing.
Sustained Use Discounts (SUDs)
SUDs are automatic discounts applied to Compute Engine VMs that run for a significant portion of the billing month. No commitment required.
- Discount begins after 25% of the month is reached.
- Maximum discount: ~30% at 100% monthly usage.
- Applies to: N1, N2, N2D, C2, C2D, M1, M2 general-purpose and memory-optimized VMs.
- Does NOT apply to: E2, Tau T2D, T2A, preemptible/Spot VMs, or App Engine flexible environment VMs.
Rightsizing
Cloud Monitoring + Recommender API surfaces VM rightsizing recommendations. The GKE Vertical Pod Autoscaler (VPA) provides container-level rightsizing. Key decision rules:
- Use Spot VMs (formerly preemptible) for fault-tolerant, batch workloads—up to ~90% cheaper than on-demand but can be preempted with a 30-second notice.
- Use E2 shared-core machines for low-traffic, development workloads.
- Use custom machine types when standard types over-provision CPU or memory.
Common Traps
- SUDs and CUDs cannot be combined on the same VM usage—CUDs take precedence and are applied first.
- Preemptible VMs have a maximum lifetime of 24 hours; Spot VMs do not have a fixed maximum lifetime but can still be preempted.
- Resource-based CUDs are not transferable between regions.
- The exam sometimes presents "reserved instances" as an option—GCP does not use that terminology; the correct term is committed use discounts.
Unlock with a paid plan
11 more sections (about 2,765 words) continue below. Every paid plan unlocks all Professional Cloud Architect study guides, the full question bank and mock exams, from $4.97/month.