Developer tools · Capacity
Cache hit ratio savings calculator.
Backend load and average latency at a given hit ratio, and the value of one more point.
Inputs
Cache effect
Enter the values.
How does cache hit ratio affect database load?
Only misses reach the backend, so backend requests per second are total requests times (1 minus the hit ratio). At a 90% hit ratio the backend sees 10% of the traffic; at 99% it sees 1%, a tenfold further reduction from a nine-point gain. Average latency in a cache-aside setup is the cache lookup time plus the miss ratio times the backend time, because every request checks the cache and only misses go on to the backend.
Worked example. 1,000 requests per second at a 90% hit ratio, 1 ms cache lookups and 50 ms backend calls: 100 requests per second reach the backend, and the average latency is 1 + 0.1 x 50 = 6 ms against 50 ms without the cache, an 8.3x average speedup.
This is expected-value arithmetic on the numbers you enter. It assumes a steady hit ratio, a uniform backend latency, and no cache stampede or warm-up period. Tail latency (p99) is dominated by misses and is not the average.
Frequently asked questions
How does hit ratio reduce database load?
Backend load is requests per second times the miss ratio. A 95% hit ratio sends 5% of requests to the database, 20 times fewer than with no cache.
How do I calculate average latency with a cache?
Average latency = cache lookup time + miss ratio x backend time, when every request checks the cache first and only misses call the backend.
Is going from 90% to 99% hit ratio worth it?
Often yes: it cuts backend traffic by ten times, since misses fall from 10% to 1%. Enter both ratios to compare the backend rate and average latency.
Why is p99 latency still high with a good hit ratio?
The slowest requests are mostly misses, so tail latency stays close to backend latency until the hit ratio is very high.
Sources
Checked
No external figures are used: the results are probability-weighted averages of the values you type in. Measure hit ratio and latencies from your own cache and database metrics. Runs entirely in your browser. Nothing you type is sent or stored.
Scope
Planning estimates, not guarantees. Real systems add overhead this model leaves out and behave differently under load. Measure on your own data and load-test before committing to a design or a bill.