Planning and Configuring a Cloud Solution
This domain tests your ability to translate business and technical requirements into the correct GCP service selections, configurations, and cost strategies before deployment. It accounts for 17% of the exam, making it one of the higher-weighted domains. Expect questions about choosing between compute, storage, and database services based on specific constraints (cost, scalability, latency, consistency), as well as questions requiring you to estimate costs or apply discount strategies.
Pricing Calculator, Sustained/Committed Use Discounts, Preemptible/Spot VMs
Google Cloud Pricing Calculator
The Google Cloud Pricing Calculator (cloud.google.com/products/calculator) is the official tool for estimating monthly costs before deploying resources. On the exam, it appears as the correct answer whenever a question asks how to estimate costs for a proposed architecture.
Sustained Use Discounts (SUDs)
Sustained Use Discounts are automatic discounts applied to Compute Engine VM usage when a VM runs for more than 25% of a billing month. No action is required — they are applied automatically.
| Usage in the Month | Approximate Discount |
|---|---|
| 25–50% | ~10% |
| 50–75% | ~20% |
| 75–100% | ~30% |
| 100% (full month) | up to 30% |
- Apply to N1, N2, N2D, C2 machine families (general-purpose and compute-optimized).
- Do not apply to E2 machine family, sole-tenant nodes, or preemptible/Spot VMs.
- Discounts are calculated on a per-vCPU and per-GB-of-RAM basis, not per-instance.
Committed Use Discounts (CUDs)
Committed Use Discounts require a 1-year or 3-year commitment in exchange for deeper discounts.
| Type | Discount vs On-Demand | Commitment |
|---|---|---|
| Resource-based CUD | Up to 57% (1yr) / 70% (3yr) | Commit to vCPU/RAM amounts |
| Spend-based CUD | Up to 25–35% | Commit to a spend level (Cloud SQL, VMware, etc.) |
- CUDs are not automatically applied; you must purchase them.
- Cannot combine CUDs with SUDs on the same resources.
- CUDs are regional — you commit per region.
Preemptible and Spot VMs
| Feature | Preemptible VM | Spot VM |
|---|---|---|
| Max runtime | 24 hours | No maximum |
| Preemption notice | 30-second shutdown warning | 30-second shutdown warning |
| Discount vs on-demand | Up to 91% | Up to 91% (similar pricing) |
| Availability | Not guaranteed | Not guaranteed |
| Can set min duration | No | No |
| Pricing model | Fixed low price | Variable (market-based) |
- Both are suitable for fault-tolerant, batch, and stateless workloads.
- Neither is appropriate for databases, stateful apps, or anything requiring uptime SLAs.
- Spot VMs are the newer model; Preemptible VMs are legacy but still tested.
Common Traps
- The exam may present "use committed use discounts for batch jobs that run unpredictably" — wrong; Spot/Preemptible is the right cost tool for intermittent workloads.
- SUDs are automatic, CUDs are manual — exam questions test whether you know which requires action.
- Preemptible VMs have a hard 24-hour limit; Spot VMs do not — this distinction appears in scenario questions.
- CUDs apply at the project level and can be shared across a billing account with commitments; don't confuse project-level with billing-account-level scope.
Worked Scenario
Requirement: A data science team runs nightly batch training jobs that take 6–8 hours and can tolerate restarts. They want the lowest possible cost. Reasoning: The jobs are fault-tolerant and batch-oriented → Preemptible or Spot VMs are correct. The jobs run less than 25% of the month, so SUDs won't apply meaningfully. CUDs are for sustained, predictable usage, not nightly batches. Answer: Use Spot VMs (or Preemptible VMs) with a checkpoint/restart mechanism.
Choosing Compute: Compute Engine vs GKE vs Cloud Run vs Cloud Functions
Decision framework diagram for selecting a compute service:
| Service | Abstraction Level | Scaling | Billing Unit | Cold Start | Best For |
|---|---|---|---|---|---|
| Compute Engine | VM (IaaS) | Manual / MIGs | Per second (min 1 min) | N/A | Legacy apps, custom OS, GPU/TPU |
| GKE | Container cluster | Node auto-scaling / HPA | Per node + cluster fee | Seconds | Microservices, stateful apps, ML training |
| Cloud Run | Serverless container | Automatic (0 to N) | Per request + CPU/memory | Milliseconds–seconds | Stateless HTTP containers, APIs |
| Cloud Functions | Serverless function | Automatic | Per invocation + compute time | Milliseconds | Event-driven, single-purpose tasks |
Key Differentiators
- Compute Engine is the only option with persistent local SSDs, GPUs, TPUs, and full OS-level access.
- GKE Autopilot removes node management; GKE Standard gives full node control. The exam tests the difference.
- Cloud Run can scale to zero (no cost when idle). It supports HTTP and gRPC and can be triggered by Pub/Sub.
- Cloud Functions Gen 2 is built on Cloud Run infrastructure and supports longer timeouts (up to 60 minutes for HTTP functions) vs Gen 1 (9 minutes).
- Cloud Functions maximum timeout: Gen 1 = 540 seconds (9 min); Gen 2 = 3,600 seconds (60 min).
Common Traps
- "Lift and shift existing VMs" → always Compute Engine, never Cloud Run or GKE without containerization.
- A question about "no infrastructure management, HTTP API, scales to zero" → Cloud Run, not Cloud Functions (Cloud Functions also scales to zero but you deploy functions, not containers).
- GKE is not serverless — you still manage nodes (unless using Autopilot).
- Cloud Functions has a max concurrent requests per instance of 1 (Gen 1) by default; Cloud Run can handle multiple concurrent requests per instance.
Worked Scenario
Requirement: A team has a containerized microservice that receives sporadic HTTP traffic. They want zero infrastructure management and want to pay only when requests are being served. Reasoning: Containerized → Cloud Run or GKE. No infra management + scale to zero + pay-per-request → Cloud Run. GKE requires managing a cluster even when idle. Answer: Cloud Run.
Unlock with a paid plan
7 more sections (about 2,431 words) continue below. Every paid plan unlocks all ACE study guides, the full question bank and mock exams, from $4.97/month.