起源故事 · 阿波罗的三套冗余
Origin Story · Apollo's Triple Redundancy
航天器、客机、核电站为什么要带一堆"用不上"的备份?因为串联系统的可靠度会被部件数量稀释 —— 一个环节失效,整个任务报废。
NASA 在阿波罗计划里把关键系统做成三重冗余:哪怕两套同时失效,还有一套能把人带回家。
这背后是一条朴素却致命的乘法律:串联是R = R₁×R₂×…(越乘越小),并联是R = 1 − (1−R₁)(1−R₂)…(失效概率相乘,越并越小)。
看懂这两个公式,就懂了为什么"多花钱做冗余"在关键系统里是必须而非奢侈。
Why do spacecraft, airliners, and nuclear plants carry stacks of backups that "never get used"? Because series reliability is diluted by the number of parts — a single failed link kills the mission.
On Apollo, NASA built the critical systems with triple redundancy: even with two of three out, the third still brought the crew home.
Behind it sits a plain but ruthless multiplication law: in series, R = R₁ × R₂ × … (every multiply shrinks reliability); in parallel, R = 1 − (1−R₁)(1−R₂) … (failure probabilities multiply downward instead).
Once those two formulas click, it's obvious why "paying extra for redundancy" in critical systems is a requirement, not a luxury.
1 可靠性框图:信号能从左流到右吗? Reliability Block Diagram: Can the Signal Flow Left to Right?
畅通Healthy绿块是正常部件、灰块是失效。串联:任意一块灰,整条路径就断;并联:只要还有一条绿色通路,系统就活着。 Green blocks are healthy components; grey blocks are failed. In series, any grey block breaks the whole path; in parallel, the system survives as long as at least one green path remains.
2 部件越多,两条曲线走向完全相反 More Parts, Two Curves That Go in Opposite Directions
红线是串联:加一个部件就乘一次,可靠度单调下降。蓝线是并联:加一个冗余就少一份失效概率,可靠度单调逼近 1。 The red line is series: each extra part is another multiply, so reliability drops monotonically. The blue line is parallel: each extra redundant unit shaves the failure probability further, so reliability climbs monotonically toward 1.
3 现实里的串/并联 Series / Parallel in the Real World
航空双备份:飞机的液压、电源、计算机常做三重冗余,单点失效也能安全落地。
Aviation triple-redundancy: hydraulics, electrical buses, and flight computers run as triple-redundant parallel chains — a single failure still leaves a safe landing on the table.
供应链多源:关键料只靠一家=串联软肋;多家供应商并联,一家断供不致命。
Multi-source supply chain: sole-sourcing a critical part is a pure series weak link; qualifying multiple suppliers in parallel turns one outage from fatal into manageable.
数据中心:双路供电、RAID 磁盘阵列、多可用区,都是用并联换高可用。
Data centers: dual power feeds, RAID arrays, multiple availability zones — every one of them buys high availability by adding parallel paths.
产线串联:多道工序串联时,整线良率是各工序良率的连乘 —— 最弱工序决定上限。
Production lines in series: when multiple stations sit back-to-back, line yield is the product of every station yield — the weakest station caps the whole line.
一句话In One Line
串联和并联是两条方向相反的乘法律:串联让可靠度相乘(越串越脆,受制于最弱一环),并联让失效概率相乘(越并越稳,逼近完美)。
上一节我们算单个部件的可靠度 R(t),这一节把它们组装成系统 —— 真实系统往往是串并混合:先把关键子系统做并联冗余,再把这些子系统串成主链。
工程师的本事,就是把钱花在最弱的串联环节上做冗余,用最小代价撬动最大可靠度。
Series and parallel are two multiplication laws pointing in opposite directions: series multiplies reliabilities (gets brittler with every extra link, bottlenecked by the weakest one); parallel multiplies failure probabilities (gets stronger with every extra path, approaching perfection).
Last page we computed R(t) for a single component; this page assembles them into a system — and real systems are almost always hybrid: critical subsystems are built parallel-redundant, then those subsystems are wired in series along the main chain.
The engineer's craft is spending the redundancy budget on the weakest series link — minimum cost, maximum reliability lift.
常见误用Common Mistakes
以为部件可靠系统就可靠。串联会稀释:95% 串 10 个只剩约 60%。
Assuming reliable parts make a reliable system. Series dilutes: ten 95% parts in series leave you with only ~60%.
冗余件共用同一故障源(同一电源/同一供应商)。真冗余要独立,否则共因失效一锅端。
Redundant units sharing a single failure source (same power feed, same supplier). True redundancy demands independence — otherwise a common-cause failure takes the whole stack out at once.
盲目堆并联。把冗余加在最弱的串联环节,性价比最高。
Piling on parallel paths everywhere. Add redundancy at the weakest series link — that's where every dollar buys the most reliability.