latencysensitive

Latencysensitive is a term used to describe systems, applications, or processes for which end-to-end latency constitutes a critical performance constraint. The term is used in engineering disciplines to describe design and operational priorities rather than a formal standard.

In latency-sensitive contexts, predictable response times and low tail latency are essential, and even small delays can degrade usability or safety.

Typical domains include high-frequency trading, real-time control systems (industrial automation, robotics), interactive online gaming, voice over IP and video conferencing, telepresence, virtual and augmented reality, and some streaming or media processing pipelines where interaction latency matters.

Key performance measures include end-to-end latency, jitter (latency variability), and tail latency distributions. Latency budgets specify maximum acceptable latency for a given operation; for some applications, budgets can be in the single-digit to tens of milliseconds, or up to a few hundred milliseconds for others.

Design approaches include reducing travel distance via edge computing and data localization, optimizing software paths, using real-time or low-latency transports, and prioritizing traffic via QoS and traffic shaping. Techniques include kernel-bypass networking, in-memory processing, hardware acceleration, and protocol choices (for example UDP-based transport with fast retransmit, minimizing round trips).

Trade-offs include higher cost, lower throughput, and potential complexity; latency sensitivity may conflict with reliability or consistency, requiring careful architectural decisions such as redundancy, caching strategies, and speculative execution.

See also latency, tail latency, QoS, edge computing.