What's changed: Created Professional Cloud DevOps Engineer Chapter 5 (Domain 5 "Optimize": APM performance collection, Active Assist insights/recommendations, recommenders (cost/security/performance/manageability/reliability); FinOps = Spot VMs/CUD/SUD/network tiers, utilization/rightsizing, observability cost, per-workload optimization for GKE/Cloud Run/Compute Engine).
5.2FinOps and optimizing workload costs
Understand FinOps practices: Spot VMs, committed-use discounts (CUD) and sustained-use discounts (SUD), network tiers, optimizing resource utilization, observability costs, leveraging recommenders, and optimizing GKE/Cloud Run/Compute Engine workload costs.
FinOps is the culture and practice where engineering and finance collaborate to continuously optimize cloud cost. Combine discounts, purchase options, and utilization to cut waste while preserving SLOs.
5.2.1Discounts, purchase options, and network tiers
| Lever | Key point |
|---|---|
| Spot VMs | Deep discount, preemptible—for fault-tolerant/batch |
| Committed-use discounts (CUD) | Commit 1/3-year usage for a discount |
| Sustained-use discounts (SUD) | Automatic discount for sustained use |
| Network tiers | Premium/Standard trade performance vs cost |
Run fault-tolerant/batch workloads that tolerate interruption on Spot VMs for deep discounts. Lower steady capacity with committed-use discounts (CUD) (1/3-year commitments); sustained-use discounts (SUD) apply automatically to long-running resources. Choose egress network tiers (Premium/Standard) to trade performance vs cost. Map "interruptible batch = Spot VMs" and "steady-use discount = CUD."
5.2.2Utilization and per-workload optimization
Raise resource utilization and trim over-allocation via rightsizing (using recommenders). Reduce observability cost with log filtering/exclusions and metric cleanup. Per workload: optimize GKE with cluster autoscaler/node pools/container rightsizing, Cloud Run via request billing and min instances tuning, and Compute Engine by combining machine type/CUD/Spot. The principle is to cut only without sacrificing SLOs. Map "measure and trim waste = utilization + rightsizing."
Common: requirement → means. E.g., "cheap interruptible batch" = Spot VMs; "discount steady use via 1/3-year commit" = CUD; "automatic discount for long use" = SUD; "egress performance/cost choice" = network tiers; "rightsize oversized instances" = rightsizing; "lower log cost" = filtering/exclusions.
Watch the mix-ups: (1) Spot VMs can be preempted—not for stateful, availability-critical prod. (2) Do not confuse CUD (pre-committed discount) with SUD (automatic sustained discount). (3) Do not breach SLOs when cutting cost (reliability trade-off).
5.2.3Section summary
- Discounts = Spot VMs (interruptible batch)/CUD (commit)/SUD (automatic); egress via network tiers
- Raise utilization; trim waste via rightsizing and observability-cost optimization (log filtering/exclusions)
- Per workload = GKE (autoscaler/node pools)/Cloud Run (min instances)/Compute Engine (machine type/CUD/Spot); preserve SLOs
Sign in to track progress — Log in.
Quick check
(just a quick review)Q1. To run interruptible batch workloads at a deep discount, which is best?
Q2. To discount steady capacity via a 1- or 3-year commitment, which is best?
Q3. Which discount applies automatically to resources used for a sustained period?
Q4. To resize over-allocated instances to match actual usage, what is this called?
Q5. To reduce observability (log) cost, which measure is best?
Q6. What constraint must you always uphold when optimizing cost?
Keep track of your progress
The full study guide is free to read. Sign up free to practice with the question bank, track what you have read, review your mistakes, and highlight passages.