Chapter 6

Scaling & Elasticity

Vertical vs horizontal, load balancers, state.

Learning objectives

Distinguish vertical and horizontal scaling
Explain elasticity vs fixed capacity in cloud billing
Plan scaling triggers for Workshop Co. seasonal demand

Scaling vocabulary

Term	Meaning	Example
Vertical scale (scale up)	Bigger single server — more RAM, CPU	Upgrade VPS 4 GB → 8 GB
Horizontal scale (scale out)	More servers sharing load	Two web VMs behind load balancer
Elasticity	Auto add/remove capacity by demand	Kubernetes HPA, cloud autoscaling groups
Fixed capacity	Pay for peak all year	Workshop Co. Proxmox sized for December gift-class rush

Workshop Co. demand pattern

Booking traffic spikes:

Saturday 7–11 AM — class registration windows
September — fall semester launch
December — gift certificate promotions

Marcus sized Proxmox for September peak — elasticity would mean adding a second web VM only in September, but that requires load balancer + shared session store. For their scale, vertical headroom on one web VM is simpler.

Worked example — when to scale out

If response time exceeds 2 seconds for 15 minutes during booking:

Short term: enable PHP opcache tuning, add RAM to VM 110 (vertical)
Medium term: clone web VM, add Traefik load balancing (horizontal)
Cloud elastic: move static assets to CDN — offload without new VMs

# Hypothetical monitoring alert — not magic autoscale
# Prometheus rule: p95_latency > 2s for 15m → page Marcus
# Manual: qm set 110 --memory 8192 && qm reboot 110

Elasticity costs

AWS autoscaling saves money for spiky unknown load. Predictable Saturday peaks favor right-sized fixed VPS — you are not paying per-minute if traffic is predictable.

Database scaling caution

PostgreSQL on VM 120 does not scale out trivially — read replicas, connection pooling (PgBouncer), or managed DB service are next steps. Workshop Co. stays single-node until connection count or disk IOPS forces change.

Premature scale-out

Two weak web servers plus load balancer add complexity before one properly sized VM is maxed. Measure first — Marcus logs Saturday CPU/RAM monthly.

Try it yourself

Choose vertical, horizontal, or CDN for each:

Nextcloud slow serving 200 MB video previews
PHP booking app CPU at 90% Saturday mornings
Marketing images on homepage

Sample answers

Vertical RAM/CPU on files VM or object storage offload (hybrid)
Vertical first (more CPU); horizontal if sustained after tuning
CDN — cache static images at edge, cheap elasticity

Check your understanding

Is elasticity the same as horizontal scaling?
Why might Workshop Co. avoid Kubernetes for 500 weekly users?

Answers

Related but not identical — elasticity is automatic add/remove; horizontal scale can be manual permanent servers.
Operational overhead exceeds benefit — Proxmox + one web VM matches team size and traffic.