Capacity
Little's Law and composed availability: how many in-flight requests a service really holds, and what a chain of dependencies does to your uptime.
L = λ × W 20 = 250/s × 0.08s
Fill in two, leave one empty and it gets solved. W is wall-clock time in the system, queueing included — which is why concurrency climbs under load even when throughput does not.
In series every dependency multiplies, so the result is always worse than the weakest link — four 99.9% services chain to 99.6%, about a day and a half a year. Parallel is redundancy: down only when every copy is down. Downtime counts a 30-day month and a 365-day year, the way SLA tables do.
Little's Law is the most useful arithmetic in capacity planning and the most frequently skipped. Concurrency equals arrival rate times latency, so a service taking 200ms at 500 requests per second is holding a hundred requests at any instant. If your pool has fifty connections, you already know what happens next.
The second half composes availability across dependencies. Services in series multiply, so four dependencies at 99.9% each give you 99.6% — worse than any single component. Redundant paths go the other way. Both directions are easy to get backwards in your head and obvious once written down.
What people open it for
- Sizing a connection or thread pool from real latency and throughput
- Showing why adding a dependency lowers the ceiling on your uptime
- Sanity-checking a load test before trusting its numbers
In the terminal
tools capacity
Every utility is also a command in the desktop's shell.