Cost Architecture#

Cost is an architecture concern, not a finance concern. Every design decision — the database engine, the compute model, the replication strategy — has a cost implication. Engineers who understand cost architecture build systems that are both performant and economically sustainable. This section covers the principles, patterns, and organisational structures for cost-aware cloud design.

Cost-Aware Architecture Principles#

The Five Principles#

Cost as a Non-Functional Requirement#

Just as you specify SLOs for latency and availability, specify cost targets:

Service: Booking API
├── Availability SLO: 99.95%
├── Latency P99: < 200ms
├── Cost target: < £0.50 per 1,000 bookings
└── Cost ceiling: £8,000/month at current volume

Architecture Trade-Offs with Cost Impact#

Every architecture decision has cost implications. The table below maps common decisions to their cost profile.

Compute Trade-Offs#

Data Trade-Offs#

Networking Trade-Offs#

Reserved Capacity Strategies#

Reserved capacity (Reserved Instances, Savings Plans, Committed Use Discounts) trades flexibility for lower unit cost.

Commitment Comparison#

Commitment Strategy#

Total compute spend
│
├── Baseline (always-on, predictable)
│   └── 70% covered by Savings Plans / CUD (1-year)
│
├── Steady growth
│   └── 20% covered by Savings Plans (flexible)
│
└── Variable/burst
    └── 10% on-demand + spot

When NOT to Commit#

Spot Architecture Patterns#

Spot instances are the most powerful cost lever in cloud computing. Designing for spot requires embracing interruption as a feature, not a failure.

Spot-Friendly Workloads#

Spot Savings Architecture#

Without Spot:
┌──────────────────────────────────────┐
│  10x m6i.xlarge (on-demand)          │
│  = 10 × £0.192/hr = £1.92/hr        │
│  = £1,382/month                      │
└──────────────────────────────────────┘

With Mixed Fleet (30% OD + 70% Spot):
┌──────────────────────────────────────┐
│  3x m6i.xlarge (on-demand)           │
│  = 3 × £0.192/hr = £0.576/hr        │
│                                      │
│  7x m6i.xlarge (spot, ~70% discount) │
│  = 7 × £0.058/hr = £0.406/hr        │
│                                      │
│  Total = £0.982/hr                   │
│  = £707/month (49% savings)          │
└──────────────────────────────────────┘

Cost Allocation Models#

Allocation Approaches#

Tagging for Cost Allocation#

Every resource must have:
├── Team: platform / bookings / search / payments
├── Environment: dev / pre / prod
├── CostCentre: CC-1234
├── Application: booking-api / search-indexer
└── ManagedBy: terraform / manual

Cost Explorer / Billing Dashboard
├── Filter by Team=bookings, Environment=prod
│   └── Shows: £12,430/month
│       ├── EC2: £5,200
│       ├── RDS: £3,800
│       ├── S3: £430
│       └── Data Transfer: £3,000
└── Filter by CostCentre=CC-1234
    └── Shows: £45,000/month (entire department)

FinOps Team Structure#

FinOps (Financial Operations) is the practice of bringing financial accountability to cloud spend. It’s a cross-functional discipline, not a finance task.

FinOps Team Model#

┌─────────────────────────────────────────────────────────────┐
│                     FinOps Practice                           │
│                                                               │
│  ┌─────────────────┐  ┌─────────────┐  ┌────────────────┐  │
│  │ Engineering      │  │ Finance     │  │ Leadership     │  │
│  │                 │  │             │  │                │  │
│  │ • Cost tools    │  │ • Budgets   │  │ • Priorities   │  │
│  │ • Architecture  │  │ • Forecasts │  │ • Trade-offs   │  │
│  │ • Optimisation  │  │ • Reporting │  │ • Investment   │  │
│  │ • Tagging       │  │ • Variance  │  │   decisions    │  │
│  └─────────────────┘  └─────────────┘  └────────────────┘  │
│                                                               │
│  Cadence: Weekly cost review, Monthly deep-dive,             │
│           Quarterly commitment planning                       │
└─────────────────────────────────────────────────────────────┘

FinOps Maturity Phases#

Unit Economics#

Unit economics measure cost efficiency in business terms, not infrastructure terms. They answer: “how much does it cost to serve one customer / process one transaction / store one GB?”

Defining Unit Metrics#

Unit Economics Dashboard#

┌─────────────────────────────────────────────────────────────┐
│                   Cost Per Booking                            │
│                                                               │
│  £0.55 ┤                                                     │
│        │     ╱╲                                              │
│  £0.45 ┤    ╱  ╲        Target: £0.50                       │
│        │   ╱    ╲─────────────────────── £0.42               │
│  £0.35 ┤  ╱                                                  │
│        │ ╱                                                   │
│  £0.25 ┤╱                                                    │
│        ├─────┬─────┬─────┬─────┬─────┬─────┬─────           │
│        Jan   Feb   Mar   Apr   May   Jun   Jul               │
│                                                               │
│  Trend: Improving (right-sizing + Graviton migration)        │
└─────────────────────────────────────────────────────────────┘

Why Unit Economics Matter#

The goal: As the business grows, total spend increases but cost per unit decreases (economies of scale). If cost per unit increases, your architecture doesn’t scale economically.

Cost Optimisation Quick Wins#

Key Takeaways#

• Treat cost as a first-class architecture requirement alongside latency, availability, and security
• Every architecture decision has a cost implication — evaluate compute, data, and networking trade-offs explicitly
• Cover baseline compute with reserved capacity (Savings Plans, CUD); use spot for burst and batch workloads
• Tag every resource for cost allocation; without tags, you cannot attribute, optimise, or budget
• FinOps is cross-functional: engineering provides optimisation, finance provides budgets, leadership provides priorities
• Measure unit economics (cost per transaction, per user, per GB) — total spend is meaningless without business context
• Cost optimisation is continuous, not a one-time project; review weekly, deep-dive monthly, plan commitments quarterly