Todos os livros

Profissional

Apps Sobre Coach Entrar Começar a ler

Quantitative Finance · Glossário

O que é Tail latency, jitter?

Também chamado de: tail latency · jitter

Definition 1.5 Low-Latency Software · Capítulo 1 — Where Latency Comes From

The tail latency of a system is the upper part of its latency distribution, reported as high percentiles: the 99th (p99), the 99.9th (p99.9), the maximum over a period. Jitter is the variability of latency from one occurrence to the next, reported as a spread (p99 minus p50, or a standard deviation); a system with low jitter is one whose tail is close to its median.

The tail of a synthetic six-stage path whose stages each stall 0.3% of the time. The sum of the stage p99s misses the stalls and understates the path’s p99; the union bound of  overstates it. Data: ll_budget.composition, 400 000 simulated messages.
Figure 1.2. The tail of a synthetic six-stage path whose stages each stall 0.3% of the time. The sum of the stage p99s misses the stalls and understates the path’s p99; the union bound of Proposition 1.6 overstates it. Data: ll_budget.composition, 400 000 simulated messages.
Per-message latency of the stages of a toy tick-to-trade path (decode one message, apply it to Book 1’s book, decide, encode an order), over 200 replays of Book 1’s 2 020-message sample, pinned to one CPU. Measured on a laptop (Intel Core Ultra 7 155H) under WSL2, no isolated cores; each figure includes the timer’s own cost of about 15\, n s. Data: bench_path.py.
Figure 1.3. Per-message latency of the stages of a toy tick-to-trade path (decode one message, apply it to Book 1’s book, decide, encode an order), over 200 replays of Book 1’s 2 020-message sample, pinned to one CPU. Measured on a laptop (Intel Core Ultra 7 155H) under WSL2, no isolated cores; each figure includes the timer’s own cost of about 15 ns15\,\mathrm{n}\mathrm{s}. Data: bench_path.py.
Ler no capítulo →