"I usually structure the abstract in four sentences: state the problem; say why it's interesting; say what your solution achieves; say what follows from it."
我通常用四句话搭摘要：说清问题；说它为何有趣；说你的方案做到了什么；说由此能得出什么。— Simon Peyton Jones,《How to Write a Great Research Paper》
本文对分布式系统中的一致性问题进行了深入的研究与探讨，首先介绍相关背景，然后分析现有方法的不足，最后提出一种新方法并进行实验验证，取得了较好的效果。跨区域数据库难以同时兼得强一致与低延迟，这直接卡住了金融应用上云。我们提出 X 协议，用因果一致加客户端缓存，把跨区读延迟从 200ms 降到 40ms，且不违反业务不变量。这说明"弱一致但够用"的中间态，比非此即彼的争论更有工程价值。前者只说"我研究了 X"，后者四句到位：问题、重要、结论含数字、启示。
This paper studies the important problem of consistency in distributed systems. We review prior work, identify limitations, propose a new approach, and show good experimental results.Cross-region databases cannot have both strong consistency and low latency, which blocks financial workloads from the cloud. We present X, which cuts cross-region read latency from 200ms to 40ms without violating business invariants. A principled "weak but sufficient" middle ground beats the all-or-nothing debate.
"Imagine that you enter a parlor. You come late. … The discussion had already begun long before any of them got there. … You listen for a while, until you decide that you have caught the tenor of the argument; then you put in your oar."
想象你走进一间客厅，你来晚了……讨论在你到来之前很久就开始了……你先听一阵，直到听懂了争论的脉络，才插进你的桨。— Kenneth Burke,《The Philosophy of Literary Form》(1941)
原理解读
综述的真正功能，是替读者画出一张"知识地图"，并标出一处空白——那处空白就是你工作的理由（gap）。没有 gap 的综述，等于没给读者一个"所以呢"。Burke 的客厅提醒你：你不是第一个说话的，也不是最后一个；贡献只有嵌进对话里才有意义。技术圈的 Related Work 同理——它不是免责声明，是定位坐标。
修改示范
关于缓存一致性，Smith(2019) 提出了 A 方法，Jones(2020) 提出了 B 方法，Lee(2021) 研究了 C。这些方法各有优缺点。缓存一致性的研究一直在"性能"与"正确性"之间摇摆：早期方案（Smith 2019）为速度牺牲了强一致；随后一派（Jones 2020）用重协议换回一致，却把延迟推高了三倍。真正没人解决的是——能不能不加协议开销，就拿到"够用的一致性"？这正是本文的切入点。清单变成一条"张力线"，末尾落在 gap 上，读者立刻知道你为何要出场。
Smith (2019) proposed A. Jones (2020) proposed B. Lee (2021) studied C.Work on cache coherence has swung between speed and correctness: early designs (Smith 2019) traded consistency for latency, while later protocol-heavy approaches (Jones 2020) restored it at 3× the cost. What no one has shown is whether "good-enough" consistency is achievable without protocol overhead — the gap this paper addresses.
"Readers do not simply read; they interpret. … The information a sentence contains is only part of what a reader gets; the other part comes from where in the sentence that information appears."
读者不是被动地读，而是主动地解读。……一个句子传递的信息，只有一部分来自内容，另一部分来自这些信息出现在句子的什么位置。— Gopen & Swan,《The Science of Scientific Writing》, American Scientist (1990)
A causal-order-based conflict detection algorithm was adopted in our system to address the data inconsistency that arises during concurrent writes.Concurrent writes cause data inconsistency. Our fix is a causal-order conflict detector that catches conflicts at write time, not after.
"Nominalizations are the zombie nouns of academic prose: they cannibalize active verbs, suck the lifeblood from adjectives and substitute abstract entities for human beings."
名词化是学术散文里的僵尸名词：它们吞噬主动动词、吸干形容词的血、用抽象实体取代活生生的人。— Helen Sword,《Zombie Nouns》, New York Times (2012)
"Never use a long word where a short one will do."
能用短词，就绝不用长词。— George Orwell,《Politics and the English Language》(1946)
The implementation of an optimization of system performance was achieved through the modification of the caching strategy.We changed the caching strategy to make the system faster.