起源故事 · Erlang 与排队论
Origin Story · Erlang and the Birth of Queuing Theory
1909 年,哥本哈根电话公司的工程师 A.K. Erlang 盯着交换机发愁:电话呼叫毫无规律地涌来,到底该配多少条线路?
他发现呼叫的到达间隔服从指数分布,由此写下排队论的奠基论文。今天你打客服听到"您前面还有 3 位",
背后就是 Erlang 一百年前的数学。指数分布,就是"下一件随机事件还要等多久"的答案。
In 1909, Copenhagen Telephone Company engineer A. K. Erlang stared at a switchboard with a budget problem: calls arrived without pattern — how many lines should he provision?
He showed that the inter-arrival times follow an exponential distribution, and wrote the founding paper of queuing theory.
Every time you hear "you are caller number 3 in the queue", you are listening to math Erlang built a century ago.
The exponential distribution answers one question: how long until the next random event?
1 无记忆:已经等过的时间,白等了 Memoryless: Time Already Waited Buys You Nothing
曲线纹丝不动curve does not budge2 随机到达:间隔忽长忽短 Random Arrivals: Gaps Stretch and Snap
每个竖线是一次事件(来电/故障)。间隔服从指数分布:短间隔最常见,偶尔出现很长的"空窗"。这正是泊松到达(上一页)的另一面。 Each tick is an event (a call or a failure). The inter-arrival times are exponential: short gaps dominate, with the occasional long quiet stretch. This is the flip side of Poisson arrivals from the previous page.
3 现实里的指数分布 Exponential in the Real World
客服排队:呼叫到达间隔。Erlang 当年的本行,也是今天呼叫中心配座席的依据。
Call-centre queues: inter-arrival times of incoming calls — Erlang's original domain, and still the textbook input for sizing agent pools today.
放射性衰变:原子核何时衰变完全无记忆,半衰期就是指数分布的体现。
Radioactive decay: a nucleus has no memory of how long it has existed. The half-life of any isotope is exponential distribution in physical form.
电子元件失效:无磨损期的随机故障(浴盆曲线的平底段)服从指数,失效率恒定。
Electronics failure: the flat middle of the bathtub curve — random failures with constant hazard rate — is exponential. MTBF is just 1/λ.
请求间隔:服务器请求、设备报修的到达间隔,排队论与容量规划的输入。
Request gaps: server requests, ticket arrivals, equipment repair calls — the raw inter-arrival input that feeds queuing models and capacity planning.
一句话In One Line
泊松数"一段时间里发生几次",指数问"下一次还要等多久" —— 同一枚硬币的两面。
它唯一的参数 λ 同时定均值(1/λ)和速率,是排队论、可靠性(MTBF)、维修策略的基石。
但记住它的致命前提:失效率恒定、不会老化。会磨损的机械零件不归它管,那是下一页的威布尔。
Poisson counts "how many events in a window"; exponential asks "how long until the next one" — two faces of the same coin.
Its lone parameter λ pins down both the mean (1/λ) and the rate, and it is the cornerstone of queuing theory, reliability work (MTBF), and maintenance policy.
Just remember the non-negotiable assumption: a constant hazard rate, no ageing. Mechanical parts that wear out belong to the Weibull on the next page, not here.
常见误用Common Mistakes
给会磨损的零件用指数(轴承、刀具)。有老化用威布尔;指数只适合恒定失效率。
Using exponential for wear-out parts (bearings, cutting tools). Use Weibull when there is ageing — exponential is only for a constant hazard rate.
混淆 λ 与平均。λ 是速率,平均等待是它的倒数 1/λ。
Confusing λ with the mean. λ is a rate; the mean wait is its reciprocal, 1/λ.
误以为"等久了就快到了"。无记忆:已等多久都不改变剩余等待分布。
Believing "I've waited a while, so it must be due soon". Memorylessness: no matter how long you have waited, the remaining-wait distribution is unchanged.