Latency

Turkish equivalent: GecikmeDomain: Performance Engineering

Elapsed time from a defined start event to a defined completion event; meaningful only when the measurement boundary is stated explicitly.

Define the Boundary First

Latency can mean network round trip, queue wait, database response, model inference or complete user-visible completion time. These are different measurements.

For a production pipeline, end-to-end latency often contains queueing, preprocessing, service execution, post-processing and transport.

Average Is Not the Tail

A low mean can coexist with rare but operationally damaging slow requests. p95/p99 and maximum behavior are therefore important when deadlines or user experience depend on the tail.

Throughput and latency are related but not interchangeable.

Updated: 01.09.2026

Related technical publications

Publications whose title or summary directly references this concept.

Tail Latency Diagnostics: Queueing, Scheduler, GC, I/O and Saturation

An engineering treatment of P95/P99 latency as an end-to-end effect of queue buildup, saturation, scheduler behavior, garbage collection, I/O and dependency chains rather than an average service-time problem.

Real-Time ASR Latency Budget: VAD, Segmentation, Queueing and Inference

A measurement model for user-visible real-time ASR latency split across capture, VAD, segmentation, queueing, inference, decoding and post-processing.

Accuracy-Latency Trade-offs in Vector Search: Recall@K, HNSW, IVF and PQ

A measurement-oriented treatment of recall, latency, memory and index-build trade-offs in approximate nearest-neighbor search across HNSW, IVF and Product Quantization.

Microprocessors

Microprocessor course notes using the 8085 and 8086 to cover registers, buses, instructions, timing, interrupts, serial/parallel I/O, DMA and memory, extended with RTOS scheduling, interrupt latency, priority inversion, cache/DMA visibility, watchdogs, and deadlines.