Amazon Web Services
Cost Optimisation & the Well-Architected Framework
Cut spend without cutting capability, and know the six pillars an architecture review is scored against.
The Well-Architected Framework is six pillars, and interviewers do ask you to name them.
A household bill. The savings are not in buying a cheaper kettle but in the immersion heater nobody remembered was on, the subscription nobody cancelled, and the standing charge nobody read.
Key Concepts
1
Operational excellence run and monitor, improve procedures
Security identity, detection, protection
Reliability recover from failure, meet demand
Performance efficiency use the right resources
Cost optimisation avoid unnecessary spend
Sustainability reduce the environmental footprint2
Pricing models, roughly in order of saving.
On-Demand no commitment, highest rate
Savings Plans commit to a $/hour for 1 or 3 years. Up to ~72%.
Compute Savings Plans apply across EC2, Fargate
and Lambda, and across instance families.
Reserved older, tied to an instance family
Spot spare capacity, up to 90% off, reclaimed with
a 2-minute warning
Graviton ARM instances, commonly 20-40% better price
for performance on a simple recompile3
The standard shape of a real bill. Steady baseline on a Savings Plan, burst on On-Demand, batch and
CI on Spot.
4
Where the money actually leaks.
idle or oversized instances rightsizing reports
unattached EBS volumes deleted instance, kept disk
old snapshots nobody owns them
NAT Gateway data processing often a surprise line item
cross-AZ traffic charged in both directions
S3 without lifecycle rules everything stays Standard forever
forgotten non-production dev running at the weekend5
S3 lifecycle rules are the easiest large saving. Move objects to Infrequent Access after 30 days, Glacier
after 90, expire after a retention period — and use Intelligent-Tiering when access patterns are unknown.
6
Tag everything and enforce it. Without team, env and service tags, Cost Explorer cannot tell anyone
which team spent what, and nothing improves because nobody owns the number.
teamenvservice
7
Measure before optimising. Compute is usually the biggest line, but data transfer and NAT charges are
where the surprises hide.
8
What the interviewer is probing.1. "Name the Well-Architected pillars." Probing: whether you know the framework by name.
Stalls: "Cost and security." Moves up: operational excellence, security, reliability,
performance efficiency, cost optimisation and sustainability.
9
2. "Where does cloud spend usually leak?" Probing: the unglamorous items. Stalls: "Oversized
instances." Moves up: unattached EBS volumes, old snapshots, NAT Gateway data processing, cross-AZ
traffic, buckets with no lifecycle rules, and non-production left running.
10
3. "Savings Plans or Spot?" Probing: matching the instrument to the workload. Stalls:
"Whichever is cheaper." Moves up: Savings Plans for the steady baseline you are confident of
running, Spot for interruptible burst and batch — usually both.
11
4. "Why does tagging matter for cost?" Probing: accountability. Stalls: "For organisation."
Moves up: without team and environment tags, Cost Explorer cannot attribute spend, and nothing
improves while nobody owns the number.