The rate at which a system, component, or network successfully completes or transfers work items or data over a specified interval, typically expressed as units per second (requests/s, MB/s, transactions/s).

Principle

Throughput is determined by the service capacity of the bottleneck resource, the offered load, and the system's efficiency; measurable as completed successful outputs per time and distinct from offered load or instantaneous bandwidth.

Demonstration

A network link reports 500 MB/s sustained throughput for file transfers; a database measures 2,000 committed transactions per second under a given workload and concurrency level.

Misapplication

Confusing throughput with offered load (sending rate) or with raw link rate; measuring peak instantaneous bursts instead of sustained completed work leads to misleading capacity conclusions.

Consequence

Accurate throughput measurement supports capacity planning, SLA definitions, and optimisation; raising throughput typically requires increasing bottleneck capacity or reducing per-item service cost.

Reversal

If throughput falls to zero the system is not delivering work (due to failure, deadlock, or collapse); unbounded attempts to increase offered load beyond bottleneck do not increase throughput and may reduce it.

Boundary

Applies to successful completed work per time for systems, subsystems, and links; excludes latency (time per item) and utilization metrics which describe resource occupation rather than completed output.

Semantic Tension

Competes conceptually with latency and utilization: higher throughput can conflict with lower latency; throughput measurements can hide tail-latency issues and fairness among flows.

Synthesis

Throughput is the measurable sustained rate of successful work completion limited by the system's bottleneck; understanding it requires measuring completed outputs, identifying bottlenecks, and reconciling trade-offs with latency and fairness.