Performance & Latency Decision Patterns

Performance & Latency Decision Patterns

Users don’t experience averages — they experience the slowest moments. Architecture decisions should focus on p95/p99 latency and perceived performance.

The Critical Path Pattern

Every system has a critical path: the chain of steps that determines user-perceived speed.

Architecture decisions should:

• Minimize work on the critical path
• Push non-essential work off the path (async)
• Cache anything expensive on the path

Decision Pattern: “Optimize Perception First”

Users care about time to first meaningful result more than total time.

Examples:

• Render a skeleton UI immediately
• Stream partial results
• Defer analytics, logging, or enrichment

Performance Budgeting

Set a budget and enforce it:

• Total page load under 2s
• API p95 under 200ms
• 99th percentile queue time < 1s

Budgets turn “performance” into enforceable constraints.

🧠 Knowledge Check

Q: Why do p95/p99 metrics matter more than averages?

<details> <summary>💡 Reveal Answer</summary>

Because users feel the slowest requests. Averages hide tail latency. If p95 is bad, many users will perceive the system as slow even if the average looks fine.

</details>

✍️ Recall Cards

• Front: The critical path is the chain that determines ______ speed. <details><summary>Back</summary>User‑perceived</details>
• Front: Optimize perception by returning ______ results early. <details><summary>Back</summary>Partial/meaningful</details>
• Front: Performance budgets make constraints ______. <details><summary>Back</summary>Explicit/enforceable</details>

🔨 Try It Yourself (Hands‑On)

Measure a critical path:

1. Pick a user flow (e.g., login)
2. List each step and estimate its latency
3. Identify one step to remove or move async
4. Propose a caching opportunity

🤔 What Would You Do?

Your analytics enrichment adds 400ms to every API response. Product wants it on every request.

<details> <summary>Best approach</summary> </details>