← Back to principles

Data

Performance at realistic load

Statement

Solutions must deliver required performance at realistic data volumes and transaction rates

What this means in practice

Performance is designed in from the start, not addressed reactively after deployment. Data models are built with performance considerations such as indexing, partitioning, and query optimisation in mind. Caching strategies are applied where appropriate, and dedicated load and performance testing is conducted before production release. Teams have access to tooling that enables them to analyse, diagnose, and tune performance throughout the solution lifecycle.

Why this matters

Systems that perform well in development but degrade under production load erode user trust and business confidence. Late discovery of performance issues is expensive to remediate and often forces architectural compromise. Designing for performance early and validating against realistic conditions reduces risk, avoids costly rework, and ensures the solution meets its service level expectations from day one.

Practices that meet this principle

  • Design data models with performance in mind, including appropriate indexing and partitioning strategies

  • If applicable, development and testing with production-representative data volumes.

  • Implement caching where access patterns justify it

  • Execute dedicated load and performance tests as part of the delivery pipeline

  • Provide teams with profiling and diagnostic tools to analyse query performance and resource utilisation

  • Define and document performance targets, including response times, throughput, and concurrency expectations

  • Performance monitoring in production - Everything before is a guess

  • Load and Stress testing is part of the CI Pipeline

Validation

A project meets this principle when:

  • Performance targets are documented with measurable thresholds

  • Testing environments contain realistic data volumes and support concurrent user simulation

  • OR:

  • No solution is released to production without evidence that it meets defined performance expectations