The elapsed time interval between when a specific data unit (packet, frame, or message) is transmitted by a sender and when that same data unit is successfully received and available for use at the destination; typically expressed as a one-way measurement and distinct from throughput or loss.

Principle

Latency is an additive temporal property: end-to-end one-way latency equals the sum of component delays (propagation, transmission, processing, queuing) along the chosen path and therefore directly limits interactive responsiveness.

Demonstration

Illustrative scenario → A real-time voice packet is timestamped at sender and observed at receiver (illustrative measurement using synchronized clocks). Recognition → The measured one-way delay is 80 ms. Action → Network operator identifies a congested queue on an intermediate router and reduces queuing by prioritizing voice traffic. Consequence → Observed one-way latency falls to 40–50 ms and perceived conversational interactivity improves.

Misapplication

Treating latency as equivalent to bandwidth or assuming low bandwidth implies high latency; or assuming one-way latency equals half the RTT without validating path symmetry or clock synchronization.

Consequence

Higher latency increases end-to-end response time, degrades interactivity for real-time applications (voice, remote control, interactive GUIs), and affects protocol performance (e.g., larger bandwidth–delay product demands larger sender windows for efficient throughput).

Reversal

When bulk throughput is the objective (large file transfer), throughput and available bandwidth govern completion time more than small reductions in latency; conversely, some applications accept higher latency if it improves robustness or spectral efficiency.

Boundary

Clearly within: measured one-way time for a packet from TX to RX. Boundary case: application-level ‘request–response’ latency that includes processing time at endpoints as well as network delay. Clearly outside: measures of channel capacity (bits/s), packet loss rate, or spectral occupancy.

Semantic Tension

Latency ↔ Throughput/Bandwidth — reducing latency can conflict with techniques (e.g., deep interleaving, heavy coding) that increase throughput or robustness, requiring trade-offs.

Synthesis

Latency is the per-packet temporal cost that determines responsiveness; it composes across network elements and must be managed alongside throughput and reliability to meet application requirements.