Compute choices
On this page 9 topics
Overview
Topics
- VMs
- containers
- Kubernetes
- serverless
Learn first
3 h of practice (estimate)
Learning outcomes
- Compare VMs, containers on managed platforms, Kubernetes, and serverless functions by control, scaling, cost, and operating effort.
- Match compute to workload shape (steady, spiky, event-driven, batch).
- Decide whether an organization should run Kubernetes at all.
Lessons
Lab and exercises
Lab
ARC-014-LAB1 · Compute decisions
Outcomes: 1–3 · Time: ~3 h
- List the Larkspur workloads: storefront API, order processing, product search indexing, nightly stock and ERP batch, image resizing.
- Describe each workload's shape (steady, spiky, event-driven, batch).
- Choose compute per workload and describe scaling behavior and operating model.
- Write each choice as an ADR including team skills and on-call capacity.
- State when each decision should be revisited.
Acceptance: the team's skills and operating capacity are part of each decision.
Exercises (case-based; six exercises)
| ID | Outcome | Task |
|---|---|---|
| ARC-014-EX01 | 1 | Compare VM, managed containers, Kubernetes, and functions for the product-image resizer on four criteria. |
| ARC-014-EX02 | 1 | List what the Larkspur team still operates when it uses managed containers versus VMs. |
| ARC-014-EX03 | 2 | Match four workloads (checkout API, Black Friday catalog reads, ERP nightly export, order-confirmation email) to compute and justify. |
| ARC-014-EX04 | 2 | A function processes a continuous stream 24 hours a day and the bill is high. Diagnose and propose a change. |
| ARC-014-EX05 | 3 | Write a one-page recommendation on whether Larkspur should run Kubernetes now. |
| ARC-014-EX06 | 3 | Define three measurable revisit triggers that would justify moving to Kubernetes. |
Show solutions
- EX01 Functions: event-triggered on upload, scale to zero, low effort, ideal; managed containers: fine but idle cost; VM: most effort; Kubernetes: overkill.
- EX02 Containers: images, deployments, autoscaling rules, application monitoring. VMs: all that plus OS patching, hardening, scaling groups, and host monitoring.
- EX03 Checkout API: managed containers, autoscaling. Catalog reads: CDN and cache plus autoscaling containers. ERP export: scheduled batch job. Email: queue-triggered function or worker.
- EX04 Per-run pricing on steady load exceeds a continuous small instance; move to a container service with steady capacity, keep functions for bursts.
- EX05 Recommend no for now: few services, small team, no platform capacity; operating cost exceeds benefit; use managed containers and revisit.
- EX06 For example: more than about 15 services, more than three teams deploying independently, a dedicated platform team of at least two, or a requirement for custom scheduling or portability.
Workplace evidence: a compute platform decision you made with operating cost considered.
Outcome coverage: all outcomes covered by exercises and the lab.
Quiz
Outcomes: O1 compare options · O2 match to workload shape · O3 decide on Kubernetes.
A self-check for you: nothing is graded, saved or sent anywhere. Choose an answer, then check it. You can try again as often as you like.
Practical part
ADR review of the ARC-014-LAB1 compute decisions plus exercises ARC-014-EX01 to EX06. Pass: quiz at least 8/10, at least five of six exercises correct, and each ADR including the team's skills and operating capacity.
Common mistakes
Kubernetes for a small team with simple workloads; serverless for long-running steady load.
Assessment
ADR review.
Workplace evidence
A compute platform decision you made with operating cost considered.
Career relevance
Cost architecture (ARC-021).