Anton Putra在 YouTube 上发了一条视频，对比了 FastAPI (Python) 和 Node.js 的服务端性能表现。第一轮是测试单纯的 Get 请求，第二轮是测试 PostgreSQL 数据库的写入。两轮测试下来，Python 的综合性能表现差不多是 Node.js 的 1/10。
一般来讲服务端的性能瓶颈都在数据库 IO，很少会遇到 CPU 瓶颈。Anton Putra 测试过不少语言和服务端框架，Python 是为数不多能在测试中段就能撞上 CPU 性能墙的。而且撞墙以后也没有恢复，后半段被锁死在 50% 的 CPU usage 上跑完了全程，也不知道为什么。
FastAPI 已经是 Python 生态内性能比较好的框架了，我真不敢想 Django 这种老乌龟得慢成什么样。没有 jit，再加上 gil 的限制，Python 的性能真的……配不上它今天的江湖地位。
最关键的是，Node.js 在 JavaScript 生态内的性能也不突出，甚至可以说是比较拉胯的……
大语言模型「懂不懂你」取决于它能获取到多少你有价值的数据。所以能无感，或低门槛的让你把有价值的数据交给它至关重要。
从这个角度看，微软、谷歌是最有潜力打造出真正基于大语言模型的生产力工具。前者有 office 这个全世界最受欢迎的办公套件，后者也有一套用户基数很大的办公套件，以及持续收集用户数据的搜索引擎和浏览器。
国内在这方面最有潜力的还是字节跳动。飞书虽然市占率不高，但是飞书已经基于豆包推出了知识问答，可以全量获取用户飞书内的所有聊天记录、文档、知识库等数据。
字节在这方面唯一的短板是不像 OpenAI 能做开放平台打通第三方服务（让腾讯和阿里把自己的数据开放给字节？这不可能），也不像 Google 本身就是一个小商业生态系统，用户能在里面满足一部分需求（现在已经打通了 Google 酒旅和 YouTube）。
腾讯也掌握了微信这个数据富矿，可惜微信里大部分都是聊天这种垃圾数据。企业微信和腾讯文档本身产品力不行，远远不如飞书好用。我甚至怀疑他们底层的数据架构也不如飞书开放灵活，想要打通、整合，可能还需要付出一点代价。
阿里有钉钉，但是我不清楚钉钉文档有多少人在用。在我的印象里，钉钉文档应该远不如飞书那么好用。而且钉钉文档的底层数据架构应该也不如飞书。现在钉钉文档 AI 也不能全量获取用户数据，还需要用户手动导入知识库，支持的格式也有限。
我们为什么会对苹果的颜色产生兴趣？是什么成就了今日的苹果？库萨克给出的答案是两个英文单词：“Jobs”（乔布斯）和“Cool”（酷）。这个答案引向苹果的另一重颜色：白色。
苹果是什么颜色的？
这只苹果，当然不是被亚当夏娃偷吃了的那只伊甸园中的禁果，也不是砸在牛顿头上帮助他发现万有引力的莫须有之果。一个名词的所指常常被时代打上深刻的烙印，甚至改变它原有的内涵。在过去的 10 年之中，当我们提到苹果的时候，绝大多数时候指的是“那个苹果”——为史蒂夫·乔布斯所创造的苹果公司及其产品。
我们为什么会对苹果的颜色产生兴趣？
回答这个问题之前，不妨先来听一个“紫牛”的故事。这个故事的讲述者，是赛斯·高汀（SethGodin），一个硅谷的实业家和市场营销专家。
故事是这样的：几年前，高汀与家人驱车穿越法国。旅程开始之时，他们都陶醉于路边以前只有在童话书中才会看到的绵延数十公里的奶牛，但 20 分钟之后，车上的人就开始对这些动物熟视无睹了，甚至，更糟糕的，他们发现，那些曾经如此美丽的奶牛已经变得令人生厌了。
“它们也许是世界上最漂亮、最吸引人的奶牛，它们也许有着超越同类的温驯的性情，以及与众不同的美丽毛皮，在阳光下的牧场上，每一头牛看起来都是如此的美丽动人。但那又如何？所有的优点都无法改变它们的长相一成不变的无趣事实。”就在这时，塞斯想到——要是此时路边出现了一头紫色的奶牛，那会怎样？
这一刻的灵光一现，催生了 21 世纪头 10 年美国出版业最畅销的一本市场营销著作——《紫牛：从默默无闻到与众不同》（Purplecow：TransformYourBusinessbyBeingRemarkable）。在这本书中，高汀断言，我们已经进入了一个消费者几乎已经拥有了所必需的一切、很难有东西能激起其消费欲望的后消费时代，传统的面向大众的产品定位和老的营销法则即将衰亡，在这个需求饱和的时代，一种产品想要卖得出去——遑论获取成功——必须是值得注意的、例外的、全新的、有趣的，换言之，成为一群平庸无奇的黑白花奶牛中的那头出类拔萃的紫牛。
什么是黑白花奶牛式的产品？什么又是紫牛式的产品？高汀给出了一个简单的特征定义：大众化无特色 VS 独一无二；向所有人发动广告攻势 VS 集中在产品成长期发动广告攻势；害怕失败 VS 害怕平淡无奇；很长的生命周期 VS 很短的生命周期；很少发生变化 VS 剧烈的变化。他特地指出，从某种意义上讲，紫牛并不比其他的牛更好，它不会生产出更多的牛奶，也不会需要更少的饲料，也没有先进到拥有多种高科技功能。它的紫色不是创新，而是一个新事物，即便它的功能仍然同市场上所有的竞争对手一样，单凭紫色这一点，就足以令它超凡脱俗。
高汀的这本书出版于 2003 年，那时，距离苹果推出 iPod 只有不到 3 年时间，令 iPod 得以执数字音乐界牛耳的 iTunesStore 服务甚至并未问世，由 iPod、iPhone 和 iPad 所掀起的苹果热潮还远远未能像今天这般，横扫在最近几年中迅速变得格外扁平的世界。正因如此，在高汀的“紫牛”群中，苹果公司只占到不算大的一个位置。
然而，8 年之后的 3 月 2 日，当前一天还在互联网上死讯纷传的史蒂夫·乔布斯出乎所有人的预料，“如一只螳螂般瘦削……也像一只螳螂般昂首阔步”地出现在旧金山芳草地艺术中心 iPad2 的发布会现场时，无需抛出那些惊人的数字——推出时间还不到一年的 iBook 网上书店，已经有超过 1 亿本书被下载；世界最大的图书出版商兰登书屋于此日正式宣布进驻，至此，全球六大图书出版商皆已成为苹果商店的合作伙伴；在全世界范围内，与信用卡绑定的苹果账户已经超过两亿个；通过在 AppStore 上销售自己开发的应用程序，全球的程序开发者们总计获得了 20 亿美元的收入；iPhone 手机的销售量达到了 1 亿台；2010 年，在推出后的 9 个月中，苹果卖出了 1500 万台 iPad，销售额高达 95 亿美元，占有了 90% 以上的平板电脑市场——只要看一眼每个人越来越苹果化的生活，便已知道，从年轻时代起便以“现实扭曲力场”著称的乔布斯，已经成功地在短短 10 年间真的扭转了现实，将一家被华尔街诊断为“病入膏肓”、市场份额滑落到历史最低点的公司，变成这个时代最闪亮耀目的一头“紫牛”。
紫牛何来？
关于苹果的颜色，多年以前，乔布斯曾经给出过一个官方定义。1983 年 10 月，麦金托什电脑揭晓软件供应商的发布会上，犹是一副年少轻狂模样的乔布斯说，“苹果是红色的，IBM 是蓝色的”。将这一红色定义演绎得最为淋漓尽致的，是那个著名的麦金托什电脑广告——一如奥威尔在《一九八四》中所预言的那个阴沉、冷寂、单调划一的未来世界，身着白色上衣、红色运动短裤的女子像一团生命之火一样跑来，挥动大锤砸碎统治人们思想的铁幕。
如果说蓝色象征的是秩序和理性，红色则是叛逆的颜色。在成长过程中深受 1968 思潮影响的乔布斯，几乎从苹果公司创立之始，就把自己放在了以 IBM 为代表的现代社会的合理化的牢笼的对面。
那是上世纪 70 年代末的美国，一个职员的国度，来自四面八方的人们像潮水一样涌进大城市，涌进高楼大厦里笼子一般的办公间。不过几年时间，大多数的人就都买得起汽车、电话和电视，中产阶级的郊区住宅也不再遥不可及。失业成了不可能的事，每个人面前摆着一条通往更快更好更高的康庄大道，仿佛只要你按照理性的社会法则修理自己，便一定可以拥抱成功与幸福。
《法国 1968：终结的开始》一书中的一段文字或许是对这种合理化生活的最佳写照：
“社会是一朵塑胶花。旧的模子，新的颜色。塑胶花永不凋谢，只是遇热融化。今儿个该发薪水了吧，我们每个月定期受贿。把我们的梦想欲望定型上模，所以呢，我，而且只是我，要往上爬。我牢牢地抓紧我的白领工作——深怕别人来抢，我默默地忍受我的蓝领劳动——毫无尊严可言。下班钟一敲，我就遁回我的蚕茧，我的欢快，我的家庭告诉我，我是对的。”
按照马克思·韦伯的观点，在这个合理化的社会中，一家企业若想获得成功，必须满足人们追逐合理化的需求，提供合理化的产品，诸如效率、可计算性、可预测性和可控制性。显然，这绝非乔布斯的苹果。
为了苛求细节上的完美而一再拖延发布时间，公司内部模糊的层级和人事关系，封闭的产品体系，嬉皮士式的企业文化……当我们回头细看苹果公司的编年史，苹果电脑时代的种种出格之处，几乎是一本 MBA 反面案例教材，而这些不按牌理出牌的叛逆属性，至今仍深烙在苹果产品的血液里。
然而，是它们令苹果成为紫牛吗？换言之，按照今天被许多苹果“粉丝”所津津乐道的这些苹果法则去打造一家新的公司，是否也能收获类似的辉煌？
事实上，在《苹果模式》（TheAppleWay）一书中，作者杰弗里·库萨克（JeffreyCruikshank）开宗明义便指出，苹果的商业模式是一种不可模仿的模式，刻意重走苹果之路的人，99% 将遭到乔布斯在 1985 年所遭受的命运——被象征着商业理性原则的董事会驱逐出局，而能够像乔布斯那样东山再起的，几乎万中无一。看一眼高汀本人的经历，便是极好的例证。尽管他似乎深谙创造紫牛的种种秘笈，但迄今为止，他的最高成就，不过是于 1998 年将自己创建的互联网市场营销软件 Yoyodyne 以 3000 万美元的价格卖给了雅虎。
是什么成就了今日的苹果？库萨克给出的答案是两个英文单词：“Jobs”（乔布斯）和“Cool”（酷）。这个答案引向苹果的另一重颜色：白色。
几乎每一款苹果产品，都会在包装盒里附送两张白色的苹果贴纸。这白色的苹果，是苹果产品追求设计、追求唯美的象征符号，更是自青年时代便热衷于禅修的乔布斯“无或一切可能”东方哲学的具体体现。
从苹果二代开始，即或人们对苹果产品有着这样或那样的诟病，但它的轻灵、简约和优雅几乎是无可置疑的。从形而上的角度观察，这形成了与以黑铁为沉重主色调的庞然大物的现代机器的巨大反差。而它所张扬的享乐主义的小众趣味，亦是对贯穿 20 世纪高效率主题的一种反讽。
然而，尽管这一白色主题一直存在于苹果的基因之中，但却直到最近 10 年，它才得到大众市场的肯定，进而成为一种席卷全球的有别于可口可乐、麦当劳和 IBM 电脑的美国文化新符号。
一种解释，可以借用英国思想家汤姆·奈仁（TomNairn）的叙述：像乔布斯这样的“伟人”，他们支配现世的权力来自某种浓稠的文化想象，一种浪漫主义的残迹，一种夸大的非理性个人主义。在一个因为审美疲劳而渴求变革的年代，这样的另类反而要比新资本主义下处处可见的庸才俗货，站在一个更有力的杠杆位置，去操纵社会的喜好。
另一种解释，听上去没有那么曼妙，但也许更接近事实：苹果的非典型性成功，并不一定代表某种新的方向，它只是人们对 20 世纪效率与理性主题的厌倦在 21 世纪的一场爆发。同青年时代的乔布斯一样，我们每个人的心中都曾经藏着一个改变世界的理想，但在绝大多数情况下，因为这样或那样的原因，我们或迟或早地选择了妥协，接受了被世界改变的命运。我们按这个社会的理性法则修剪自己的枝丫，因为这样才能让成功——或者是不失败——的可能性最大化。但是，在我们的心里，依然藏着一个乔布斯式的自我，追求完美，不随波逐流，但却创造了潮流。
在这种意义上，乔布斯，这个符号式的人物，成为苹果特质的一个背书。不可否认的是，随着苹果变得越来越大，它的颜色也在慢慢地发生着微妙的变化——事实上，它越是成功，就离人们的想象越远。将流水线的高效率发挥到极致的组装工厂，保证了 iPad 令竞争对手无可企及的低成本。每个人都要玩《愤怒的小鸟》和《植物大战僵尸》，否则便似乎被时代的潮流所抛弃。它越来越多地散发着商业社会的金属光泽，变成一只银色的苹果。生活在消费社会中的我们以为可以凭借苹果式的产品摆脱从一个理性系统转向另一个理性系统的惯性路线，但真相是，在这个全球现代化的时代，王小波的预言从未像今天这般真实：“白银时代的人蒙神恩宠，终身不会衰老，也不会为生计所困。他们没有痛苦，没有忧虑，一直到死，相貌和心灵都像儿童。死掉以后，他们的幽灵还会在尘世上游荡。”
但是，只要乔布斯在那里，因为他的强大到无可比拟的“现实扭曲力场”，我们仿佛就可以说服自己，这只苹果，是与众不同的，热爱这只苹果的我们，是与众不同的。
关于乔布斯，最新的消息是，他终于要出自传了。
西蒙舒斯特出版公司上周宣布，由曾任 CNN 首席执行官和《时代》周刊执行主编的瓦尔特·伊萨克森（WalterIsaacson）执笔撰写的《iSteve：TheBookofJobs》，将于 2012 年初问世。在此之前，尽管市面上流传着十余本关于乔布斯的传记，但它们都未能得到乔布斯的正式授权。甚至于其中一本的出版商还为此在 2005 年遭到苹果在线商店的杯葛。
这一次的情形显然不同。据西蒙舒斯特的发言人称，从 2009 年开始，乔布斯本人和他的家人、朋友、同事就向伊萨克森一一敞开了大门。这让人无法不联想起乔布斯曾经多次说过的一句话：“现在还不到回忆的时候……一个人死之前不应出版自传，以免活着后悔。”
那么，现在，是时候了吗？■
版权声明：凡注明“三联生活周刊”、“爱乐”或“原创”来源之作品（文字、图片、音频、视频），未经三联生活周刊或爱乐杂志授权，任何媒体和个人不得转载、链接、转贴或以其它方式使用；已经本刊、本网书面授权的，在使用时必须注明“来源：三联生活周刊”或“来源：爱乐”。违反上述声明的，本刊、本网将追究其相关法律责任。
所有基于 Electron 开发的软件似乎都会在 macOS Tahoe 遇到水土不服的问题，包括但不限于卡顿、发热。Electron 是最常用的跨平台开发框架，也就是说，包括 VSCode、Cursor 在内的很多常用软件都会中招。
成因似乎是 macOS Tahoe 的 WindowServer 组件有 Bug，在渲染带阴影窗口时存在严重性能回退，所有带阴影 Electron 窗口会异常消耗 GPU 资源。
暂时没有看到关于 Chrome 的反馈，但是我感觉 Chrome 在升级后资源开销也变大了。
苹果 💊。
毫无意外，macOS Tahoe 和 iOS 26 就是一次计划报废。新增加的毛玻璃特效在当下这个时点看不出交互上的必要性，反而会持续消耗计算机的算力。我观察了一下我 MacBook 的 GPU 用量，此前闲时用量很少超过 20%，升级到 Tahoe 后，闲时大概 30% 左右，CPU 闲时用量此前很少能超过 20%，现在也来到了 30%。苹果这边是真没活了，就想通过这种方式催促用户换新设备。
DALL-E/Every illustration.
_The internet is like water—we take its existence for granted, but its birth was by no means pre-ordained. A constellation of inventors, organizations, and efforts all contributed to its creation. In one of her signature deep dives, Contrary writer Anna-Sofia Lesiv excavates the history of digital communication infrastructure, from the invention of the telephone to the widespread installation of fiber-optic cable and big tech’s subsidization of undersea cables. Read this chronicle to understand how the internet’s decentralized origins led to its current state as fractured spaces governed by private entities—and its implications for its future accessibility. —__Kate Lee_
The internet is a universe of its own. For one, it codifies and processes the record of our society’s activities in a shared language, a language that can be transmitted across electric signals and electromagnetic waves at light speeds.
The infrastructure that makes this scale possible is similarly astounding—a massive, global web of physical hardware, consisting of more than 5 billion kilometers of fiber-optic cable, more than 574 active and planned submarine cables that span a over 1 million kilometers in length, and a constellation of more than 5,400 satellites offering connectivity from low earth orbit (LEO).
According to recent estimates, 328 million terabytes of data are created each day*. _There are billions of smartphone devices sold every year*, _and although it’s difficult to accurately count the total number of individually connected devices, some estimates put this number between 20 and 50 billion.
“The Internet is no longer tracking the population of humans and the level of human use. The growth of the Internet is no longer bounded by human population growth, nor the number of hours in the day when humans are awake,” writes Geoff Huston, chief scientist at the nonprofit Asia Pacific Network Information Center.
But without a designated steward, the internet faces challenges for its continued maintenance—and for the accessibility it provides. These are incredibly important questions. But in order to grasp them, it’s important to understand the internet in its entirety, from its development to where we are today.
Unlock the power of algorithmic trading with Composer. Our platform is designed for both novices and seasoned traders, offering tools to create automatic trading strategies without any coding knowledge.
With more than $1 billion in trading volume already, Composer is the platform of choice. Create your own trading strategies using AI or choose from over 1,000 community-shared strategies. Test their performance and invest with a single click.
In the analog era, every type of data had a designated medium. Text was transmitted via paper. Images were transmitted via canvas or photographs. Speech was communicated via sound waves.
A major breakthrough occurred when Alexander Graham Bell invented the telephone in 1876. Sound waves that were created on one end of the phone line were converted into electrical frequencies, which were then carried through a wire. At the other end, those same frequencies were reproduced as sound once again. Speech could now transcend physical proximity.
Unfortunately, while this system extended the range of conversations, it still suffered from the same drawbacks as conversations held in direct physical proximity. Just as background noise makes it harder to hear someone speak, electrical interference in the transfer line would introduce noise and scramble the message coming across the wire. Once noise was introduced, there was no real way to remove it and restore the original message. Even repeaters, which amplified signals, had the adverse effect of amplifying the noise from the interference. Over enough distance, the original message could become incomprehensible.
Still, the phone companies tried to make it work. The first transcontinental line was established in 1914, connecting customers between San Francisco and New York. It comprised 3,400 miles of wire hung from 130,000 poles.
In those days, the biggest telephone provider was the American Telephone and Telegraph Company (AT&T), which had absorbed the Bell Telephone Company in 1899. As long-distance communications exploded across the United States, Bell Labs, an internal research department of electrical engineers and mathematicians, started to think about expanding the network’s capacity. One of these engineers was Claude Shannon.
In 1941, Shannon arrived at Bell Labs from MIT, where the ideas behind the computer revolution were in their infancy. He studied under Norbert Wiener, the father of cybernetics, and worked on Vannevar Bush’s differential analyzer, a type of mechanical computer that could resolve differential equations by using arbitrarily designed circuits to produce specific calculations.
Source: __Computer History Museum_._
It was Shannon’s experience with the differential analyzer that inspired the idea for his master’s thesis. In 1937, he submitted “A Symbolic Analysis of Relay and Switching Circuits.” It was a breakthrough paper that pointed out that boolean algebra could be represented physically in electrical circuits. The beautiful thing about these boolean operators is that they require only two inputs—on and off.
It was an elegant way of standardizing the design of computer logic. And, if the computer’s operations could be standardized, perhaps the inputs the computer operated on could be standardized too.
When Shannon began working at Bell Labs during the Second World War, in part to study cryptographic communications as part of the American war effort, there was no clear definition of information. “Information” was a synonym for meaning or significance, its essence was largely ephemeral. As Shannon studied the structures of messages and language systems, he realized that there was a mathematical structure that underlied _information. _This meant that information could, in fact, be quantified. But to do so, information would need a unit of measurement.
Shannon coined the term “bit” to represent the smallest singular unit of information. This framework of quantification translated easily to the electronic signals in a digital computer, which could only be in one of two states—on or off. Shannon published these insights in his 1948 paper, “A Mathematical Theory of Communication,” just one year after the invention of the transistor by his colleagues at Bell Labs.
The paper didn’t simply discuss information encoding. It also created a mathematical framework to categorize the entire communication process in this way. For instance, Shannon noted that all information traveling from a sender to a recipient must pass through a channel, whether that channel be a wire or the atmosphere.
Shannon’s transformative insight was that every channel has a threshold—a maximum amount of information that can be delivered reliably to a sender. As long as the quantity of information carried through the channel fell below the threshold, it could be delivered to the sender intact, even if noise had scrambled some of the message during transmission. He used mathematics to prove that any message could be error-corrected into its original state if it traveled through a large-enough channel.
The enormity of this revolution is difficult to communicate today, mainly because we’re swimming in its consequences. Shannon’s theory implied that text, images, films, and even genetic material could be translated into his informational language of bits. It laid out the rules by which machines could talk to one another—about anything.
At the time that Shannon developed his theory, computers could not yet _communicate _with one another. If you wanted to transfer information from one computer to the other, you would have to physically walk over to the other computer and manually input the data yourself. However, talking machines were now an emerging possibility. And Shannon had just written the handbook for how to start building it.
The telephone system was the only interconnected network by the mid-20th century. AT&T was the largest telephone network at the time. It had a monstrous continental web with hanging copper wires criss-crossing across the continent.
The telephone network worked primarily through circuit switching. Every pair of callers would get a dedicated “line” for the duration of their conversation. When it ended, an operator would reassign that line to connect other pairs of callers, and so on.
At the time, it was possible to get computers “on the network” by converting their digital signals into analog signals, and sending the analog signals through the telephone lines. But reserving an entire line for a single computer-to-computer interaction was seen as hugely wasteful.
Leonard Kleinrock, a student of Shannon’s at MIT, began to explore the design for a digital communications network—one that could transmit digital bits instead of analog sound waves.
His solution, which he wrote up as his graduate dissertation, was a packet-switching system that involved breaking up digital messages into a series of smaller pieces known as packets. Packet switching shared resources among connected computers. Rather than having a single computer’s long communiqué take up an entire line, that line could instead be shared among several users’ packets. This design allowed more messages to get to their destinations more efficiently.
For this scheme to work, there would need to be a network mechanism responsible for granting access to different packets very quickly. To prevent bottlenecks, this mechanism would need to know how to calculate the most efficient, opportunistic path to take a packet to its destination. And this mechanism couldn’t be a central point in the system that could get stuck with traffic—it would need to be a distributed mechanism that worked at each node in the network.
Kleinrock approached AT&T and asked if the company would be interested in implementing such a system. AT&T rejected his proposal—most demand was still in analog communications. Instead, they told him to use the regular phone lines to send his digital communications—but that made no economic sense.
“It takes you 35 seconds to dial up a call. You charge me for a minimum of three minutes, and I want to send a hundredth-of-a-second of data,” Kleinrock said.
It would take the U.S. government to resolve this impasse and command such a network into existence. In the late 1960s, shaken by the Soviet Union’s success in launching Sputnik into orbit, the U.S. Department of Defense began investing heavily in new research and development. It created ARPA, the Advanced Research Projects Agency, which funded various research labs across the country.
Robert Taylor, who was tasked with monitoring the programs’ progress from the Pentagon, had set up three separate Teletype terminals for each of the ARPA-funded programs. At a time when computers cost anywhere from $500,000 to several million dollars, three computers sitting side-by-side seemed like a tremendous waste of money.
“Once you saw that there were these three different terminals to these three distinct places the obvious question that would come to anyone's mind [was]: why don't we just have a network such that we have one terminal and we can go anywhere we want?” Taylor asked.
This was the perfect application for packet switching. Taylor, familiar with Kleinrock’s work, commissioned an electronics company to build the types of packet switchers Kleinrock had envisioned. These packet switchers were known as interface message processors (IMPs). The first two IMPs were connected to mainframes at UCLA and Stanford Research Institute (SRI), using the telephone service between them as the communications backbone. On October 29, 1969, the first message between UCLA and SRI was sent. ARPANET was born.
ARPANET grew rapidly. By 1973, there were 40 computers connected to IMPs across the country. As the network grew faster, it became clear that a more robust packet-switching protocol would need to be developed. ARPANET’s protocol had a few properties that prevented it from scaling easily. It struggled to deal with packets arriving out of order, didn’t have a great way to prioritize them, and lacked an optimized system to deal with computer addresses.
Source: __Computer History Museum_._
By 1974, researchers Vinton Cerf and Robert Khan came out with “A Protocol for Packet Network Intercommunication.” They outlined the ideas that would eventually become Transmission Control Protocol (TCP) and Internet Protocol (IP)—the two fundamental standards of the internet today. The core idea that enabled both was a “datagram,” which wrapped the packets in a little envelope. That envelope would act as a little header at the front of each packet that would include the address it was going to, along with other helpful bits of info.
In Cerf and Khan’s conception, the TCP would run on the end-nodes of the network—meaning that it wouldn’t run on the routers and obstruct traffic, but instead on users’ computers. The TCP would do everything from breaking messages into packets, placing the packets into datagrams, ordering the packets correctly at the receiver’s end, and performing error correction.
Packets would then be routed via IP through the network, which ran on all the packet-directing routers. IP only looked at the destination of the packet, while remaining entirely blind to the contents it was transmitting, enabling both speed and privacy.
These protocols were trialed on a number of nodes within the ARPANET, and the standards for TCP and IP were officially published in 1981. What was exceedingly clever about this suite of protocols was its generality. TCP and IP did not care which carrier technology transmitted its packets, whether it be copper wire, fiber-optic cable, or radio. And they imposed no constraints on what the bits could be formatted into—video text, simple messages, or even web pages formatted in a browser.
_Source: __David D. Clark,_ Designing an Internet.
This gave the system a lot of freedom and potential. Every use case could be built and distributed to any machine with an IP address in the network. Even then, it was difficult to foresee just how massive the internet would one day become.
David Clark, one of the architects of the original internet, wrote in 1978 that “we should … prepare for the day when there are more than 256 networks in the Internet.” He now looks upon that comment with some humor. Many assumptions about the nature of computer networking have changed since then, primarily the explosion in the number of personal computers. Today, billions of individual devices are connected across hundreds of thousands of smaller networks. Remarkably, they all still do so using IP.
Although ARPANET was decommissioned in 1986, the rest of the connected computers kept going. Residences with personal computers used dial-up to get email access. After 1989, a new virtual knowledge base was invented with the World Wide Web.
With the advent of the web, new infrastructure, consisting of web servers, emerged to ensure the web was always available to users, and programs like web browsers allowed end nodes to view the information and web pages stored in the servers.
Source: __Our World in Data_._
As the number of connected people increased in hockey-stick fashion, carriers finally began realizing that dial-up—converting digital to analog signals—was not going to cut it anymore. They would need to rebuild the physical connectivity layer by making it digital-first.
The single biggest development that would enable this and alter the internet forever was the mass installment of fiber-optic cable throughout the 1990s. Fiber optics use photons traveling through thin glass to increase the speed of information flow. The fastest connection possible with copper wire was about 45 million bits per second (mbps). Fiber optics made that connection more than 2,000 times faster. Today, residences can hook into a fiber-optic connection that can deliver them 100 billion bits per second (gbps).
Fiber was initially laid down by telecom companies offering high-quality cable television service to homes. The same lines would be used to provide internet access to these households. However, these service speeds were so fast that a whole new category of behavior became possible online. Information moved fast enough to make applications like video calling or video streaming a reality.
The connection was so good that video would no longer have to go through the cable company’s digital link to your television. It could be transmitted through those same IP packets and viewed with the same experience on your computer.
YouTube debuted in 2004 and Netflix began streaming in 2007. The data consumption of American households skyrocketed. Streaming a film or a movie requires about 1 to 3 gigabytes of data per hour. In 2013, the median household consumed 20-60 gigabytes of data per month. Today, that number falls somewhere about 587 gigabytes.
And while it may have been the government and small research groups that kickstarted the birth of the internet, its evolution henceforth was dictated by market forces, including service providers that offered cheaper-than-ever communication channels and users that primarily wanted to use those channels for entertainment.
If the internet imagined by Cerf and Kahn was a distributed network of routers and endpoints that shared data in a peer-to-peer fashion, the internet of our day is a wildly different beast.
The biggest reason for this is that the internet today is not primarily used for back-and-forth networking and communications—the vast majority of users treat it as a high-speed channel for content delivery.
In 2022, video streaming comprised nearly 58 percent of all Internet traffic. Netflix and YouTube alone accounted for 15 and 11 percent, respectively.
This even shows up in internet service provision statistics. Far more capacity is granted for downlink to end nodes than for uplink—meaning there is more capacity to provide information to end-user nodes than to send data through networks. Typical cable speeds for downlink might reach over 1,000 mbps, but only about 35 mbps are granted for uplink. It’s not really a two-way street anymore.
Even though the downlink speeds enabled by fiber were blazingly fast, the laws of physics still imposed some harsh realities for global internet companies with servers headquartered in the United States. The image below shows the “round-trip time” for various global users to connect to Facebook in 2011.
Source: __Geoff Huston_._
At the time, Facebook users in Asia or Africa had a completely different experience to their counterparts in the U.S. Their connection to a Facebook server had to travel halfway around the world, while users in the U.S. or Canada could enjoy nearly instantaneous service. To combat this, larger companies like Google, Facebook, Netflix, and others began storing their content physically closer to users through CDNs, or “content delivery networks.”
These hubs would store caches of the websites’ data so that global users wouldn’t need to ping Facebook’s main servers—they could merely interact with the CDNs. The largest companies realized that they could go even further. If their client base was global, they had an economic incentive to build a global service infrastructure. Instead of simply owning the CDNs that host your data, why not own the literal fiber cable that connects servers from the United States to the rest of the world?
In the 2020s, the largest internet companies have done just that. Most of the world’s submarine cable capacity is now either partially or entirely owned by a FAANG company—meaning Facebook (Meta), Amazon, Apple, Netflix, or Google (Alphabet). Below is a map of some of the sub-sea cables that Facebook has played a part in financing.
Source: __Telegeography_._
These cable systems are increasingly impressive. Google, which owns a number of sub-sea cables across the Atlantic and Pacific, can deliver hundreds of terabits per second through its infrastructure.
In other words, these applications have become so popular that they have had to leave traditional internet infrastructure and operate their services within their own private networks. These networks not only handle the physical layer, but also create new transfer protocols —totally disconnected from IP or TCP. Data is transferred on their own private protocols, essentially creating digital fiefdoms.
This verticalization around an enclosed network has offered a number of benefits for such companies. If IP poses security risks that are inconvenient for these companies to deal with, they can just stop using IP. If the nature by which TCP delivers data to the end-nodes is not efficient enough for the company’s purposes, they can create their own protocols to do it better.
On the other hand, the fracturing of the internet from a common digital space to a tapestry of private networks raises important questions about its future as a public good.
For instance, as provision becomes more privatized, it is difficult to answer whose shoulders the responsibility of providing access to the internet as a “human right,” as the U.N. describes, will fall on.
And even though the internet has become the de facto record of recent society’s activities, there is no one with the dedicated role of helping maintain and preserve these records. Already, the problem known as link rot is beginning to affect everyone from the Harvard Law Review, where, according to Jonathan Zittrain, three quarters of all links cited no longer function. This occurs even at The New York Times, where roughly half of all articles contain at least one rotted link.
The consolation is that the story of the internet is nowhere near over. It is a dynamic and constantly evolving structure. Just as high-speed fiber optics reshaped how we use the internet, forthcoming technologies may have a similarly transformative effect on the structure of our networks.
SpaceX’s Starlink is already unlocking a completely new way of providing service to millions. Its data packets, which travel to users via radio waves from low earth orbit, may soon be one of the fastest and most economical ways of delivering internet access to a majority of users on Earth. After all, the distance from LEO to the surface of the Earth is just a fraction of the length of subsea cables across the Atlantic and Pacific oceans. Astranis, another satellite internet service provider that parks its small sats in geostationary orbit, may deliver a similarly game-changing service for many. Internet from space may one day become a kind of common global provider. We will need to wait and see what kind of opportunities a sea change like this may unlock.
Still, it is undeniable that what was once a unified network has, over time, fractured into smaller spaces, governed independently of the whole. If the initial problems of networking involved the feasibility of digital communications, present and future considerations will center on the social aspects of a network that is provided by private entities, used by private entities, but relied on by the public.
Anna-Sofia Lesiv is a writer at venture capital firm __Contrary_, where she originally published __this piece_. She graduated from Stanford with a degree in economics and has worked at Bridgewater, Founders Fund, and 8VC.
最近两天在学 Astro。
Astro 的一大优势就是能静态生成网站。通过编译的方式，将 JavaScript 写的网站编译成只包含 HTML 和 CSS 的静态文件。不管是上到服务器用 Nginx 转发，还是放到静态网站托管平台，还是直接放 CDN，都能运行，不挑环境。
编译的好处就是能一处编译，处处运行。不依赖运行时和解释器。
现在 JavaScript 的运行时这么多，为啥没有人用 golang 写一个？然后继承 Golang 可以直接编译为二进制的优势，让 JavaScript 也能编译为二进制，做到一处编译，处处运行。
Ai 要真正提升生产力，做出真正不一样的产品，还是需要和已有的数据整合起来。
从这一点来看谷歌的优势太大了。目前谷歌的 Ai Workspace 是整合最好的 Ai office 产品。微软的 copilot 365 输在了 office 三件套用户体验不行。notion ai 能整合的资源有限，价格没有竞争力。wps 的用户基本不上云，金山没有数据可用。国内最有希望的还是字节，飞书体验没有问题，就看字节能把豆包优化成什么样了。
Open Ai、anthropic 起了个大早，赶了个晚集。如果 chrome 真的能被剥离，这俩公司谁能买下来，谁就有一线生机。
历史通常被描述为国王、战争和胜利者的历史，但真正的历史其实是这些思想和技术的历史——无论是车轴、印刷机还是抄本。 所以如果你把计算机放在这样的传承中，把它当作之前一切的美丽演化，那么对我来说，这种演化的关键应用就是我们能够获取更多的人类潜能。
「在以電腦爲媒介的世界裏，幾乎所有人都是文盲。她們是被剝奪了公民權的底層，無法參與塑造自己所在的世界。」——Bret Victor, 2022
这里的文盲，应该指的是，不会编程。
