Cache Hit Ratio Economics
The non-linear relationship between cache hit rate and backend load, which determines whether a cache improvement is worth making.
The number that matters is not the hit rate but the miss rate, because misses are what the backend sees.
Going from 50% to 75% hit rate halves backend load. Going from 90% to 95% halves it again. Going from 95% to 99% divides it by five. Each apparently small improvement near the top is worth more than a large improvement in the middle — and this is why teams that dismiss "only a 4% gain" are frequently leaving the biggest win on the table.
It also explains the reverse: a cache hit rate falling from 99% to 95% multiplies backend load by five, which is enough to take down a database that was comfortable. A cache that becomes load-bearing must be capacity-planned as such, and the origin must be able to survive a complete cache loss — a flush, a failover, a cold start — or the cache is not an optimisation but an undeclared dependency.
Levers on hit rate: longer TTLs (traded against staleness), a larger cache (the working set may be just above capacity — this is often the cheapest fix), better key design, and cache warming after deployments to avoid a cold-start thundering herd.