● Interactive · IH02 scrollytelling

From 800ms to 190ms

A caching story you scroll through. The chart stays pinned; each step changes one thing — the way a good explainer reveals a system part by part.

↓ scroll to advance

budget 200ms
800ms p99
Baseline
01

Every request hit the database

No cache. p99 latency sat at 800ms — the hot path paid full price every time.

02

Add a cache, keyed by request

Repeat reads now skip the database. p99 falls to ~300ms — but we're not done.

03

Cold caches spiked on deploy

Every deploy emptied the cache, so the first wave of traffic spiked back up.

04

Warm the cache on boot

Prime it before traffic arrives. p99 settles at 190ms — under budget, and stable.