Part 5 · 1 chapters · ~8 min

Cost Engineering

Tagging and allocation, unit economics per request, transfer and customer, the usual surprises (logs, data transfer, idle capacity, vendor calls), rightsizing and autoscaling, storage tiers, reserved and spot capacity, cost in design reviews, and FinOps habits.

10

Unit costs, owned by teams

code
unit cost = monthly cost of the service / business units served
          = $16,200 / 1,800,000 transfers ≈ $0.009 per transfer

design review question: "What does this add per transfer at 10× volume?"
  + one extra KYC lookup per transfer at $0.02           → doubles the unit cost: cache results for 24 h
  + DEBUG logging on the hot path at 2 KB × 1.8M × 30    → ~108 GB/month of logs: sample at 1%
  + cross-AZ chatter between services                    → data transfer charges: co-locate or batch
levertypical effect
rightsizing (CPU and memory requests near p95 usage)large savings on over-provisioned clusters
autoscaling and scale-to-zero for dev and batchpay for use, not peaks
commitments (savings plans, reserved) for the steady basediscount on predictable load
spot or preemptible for interruptible workdeep discounts for batch and CI
storage tiers and lifecycle rulesmove cold data to cheaper classes automatically
UNIT COSTS FOR ONE SERVICE (ILLUSTRATIVE)
monthly bill divided by what the business sells
compute$4,200database$3,100logs and metrics$2,600data transfer$900third-party APIs$5,400
swipe the figure sideways, or tap expand for full screen
1/4
the bill
A raw monthly bill says little. Break it down by category and owner using tags (team, service, environment).
tag everythingteam, service, env