Definition
The elapsed time between the initiation of an operation (or request) and the moment its first or relevant result is observed by the requester in a given system or communication path.

Principle

Principle
Latency is governed by the sum of propagation, transmission, serialization, queuing, processing, and protocol-induced delays along the path between initiator and responder.

Demonstration

Demonstration
A client sends an HTTP GET to a server and measures the time from sending the TCP packet to receiving the first bytes of the HTTP response; that elapsed interval is the observed request latency for that transaction.

Misapplication

Misapplication
Treating latency as identical to throughput or bandwidth (higher throughput does not necessarily reduce single-request latency), or assuming a single measurement represents all future latency behavior without accounting for variance or outliers.

Consequence

Consequence
When accurately measured and minimized where appropriate, low latency improves responsiveness for interactive applications (e.g., real-time control, voice/video calls, trading systems); understanding the components of latency guides targeted optimizations.

Reversal

Reversal
Viewing latency as the opposite of capacity (which emphasizes total volume handled) — reversing focus to aggregate throughput can mask per-request delays important to users.

Boundary

Boundary
Applies to first/initial relevant response time; excludes end-to-end completion time metrics like total transfer time unless explicitly defined as such. Measurements depend on test points, clock synchronization, and sampling method; network-only latency excludes application-level processing unless specified.

Semantic Tension

Semantic Tension
Latency is often confused with jitter (variability of latency) and with throughput/bandwidth (rate of data transfer); distinguishing them is necessary for correct diagnostics and optimization.

Synthesis

Synthesis
Latency is the time-cost of initiating and receiving a first meaningful response in a system; it arises from identifiable physical and queuing causes, must be measured at defined points with uncertainty acknowledged, and is distinct from rate-based properties like throughput.