Cost and Performance Optimization

Cost and Performance Optimization

Systematically identify and resolve cost inefficiencies and performance bottlenecks in running systems. Apply FinOps principles and performance engineering practices to deliver measurable improvements.

Overview

Optimisation without measurement is guessing. This skill starts with data collection, identifies the most impactful opportunities, and produces a prioritised improvement plan with expected outcomes.

Cost Optimization (FinOps)

Step 1: Cost Analysis

Identify the cost breakdown by:

• Service/component (which is the largest cost driver?)
• Resource type (compute, storage, network, database, egress)
• Environment (production vs. non-production)
• Time pattern (constant vs. spiky — opportunity for savings?)

Step 2: Common Cost Opportunities

Opportunity	Typical Saving	Effort	Risk
Right-size over-provisioned compute	30-50% compute cost	Low	Low
Reserved instances / Savings Plans (1-year)	30-40% vs on-demand	Low	Low
Delete unused resources (orphaned volumes, snapshots, LBs)	Variable	Low	Low
Reduce non-prod environments operating hours	60-70% non-prod compute	Low	Low
Move infrequent-access data to cheaper storage tiers	50-80% storage cost	Medium	Low
Cache frequently read data (reduce DB reads)	20-40% DB cost	Medium	Medium
Optimise network egress (CDN, compression)	Variable	Medium	Low

Step 3: Produce Cost Optimisation Report

# Cost Optimisation Report: [Service/System]
Analysis Period: [YYYY-MM to YYYY-MM]
Current Monthly Cost: [£/$]

## Top 5 Opportunities
| Opportunity | Current Cost | Estimated Saving | Effort | Recommended Action |
|---|---|---|---|---|
| [item] | [£/$] | [£/$ or %] | S/M/L | [specific action] |

## Quick Wins (implement this sprint)
[List of low-effort, low-risk actions with expected savings]

## Medium-Term Actions (next quarter)
[Actions requiring more planning]

Performance Optimization

Step 1: Identify Bottlenecks

Use the performance data available:

• Latency: which endpoints or operations are slowest? (p95, p99 distribution)
• Throughput: where does throughput degrade under load?
• Resource saturation: which resource hits 100% first (CPU, memory, DB connections)?
• Error rate under load: does error rate increase at higher traffic?

Step 2: Query and Database Optimization

For slow database queries:

1. Identify queries by execution time (EXPLAIN ANALYZE)
2. Check for missing indexes on WHERE and JOIN columns
3. Check for N+1 query patterns (load relationship once, not per record)
4. Check for unbounded queries (missing LIMIT clause)
5. Evaluate query caching (Redis/Memcached) for read-heavy, stable data

Step 3: Application Performance

• Identify synchronous operations that should be async (move to queue)
• Identify repeated computation that should be cached
• Identify excessive serialisation/deserialisation
• Identify memory leaks (growing heap over time)

Step 4: Performance Improvement Plan

# Performance Improvement Plan
Baseline: [p95 latency: Xms, throughput: Y rps, error rate: Z%]
Target: [p95 latency: Xms, throughput: Y rps]

## Prioritised Improvements
| # | Improvement | Expected Gain | Effort | Owner |
|---|---|---|---|---|
| 1 | [improvement] | [X% latency reduction] | S/M/L | [role] |

Additional Resources

• references/finops-framework.md — FinOps principles and cloud-specific optimisation
• references/performance-profiling.md — Profiling techniques per tech stack