Don't be fooled by serverless

原文英文,约600词,阅读约需2分钟。发表于:

Cloud aficionados love pinning the true promise of the cloud on serverless functions and services. Not getting the savings you thought you would with the cloud? It's because you didn't go serverless. Frustrated with the complexity of the cloud? Serverless! Performance questions? SERVERLESS! Serverless has become a mantra to chant because it's still appears just magic enough that most people won't question the fundamentals. But you should.Let's start at the beginning. Clouds, and VPS's before that, work on the age-old principle of buying in bulk and selling by the piece. You run one big server for $1,000/month, then you rent it out to seven people for $200/month, and voila, you've cleared a $400/month profit.That actually works well for all parties, if the seven customers aren't too taxing on the equipment or if they tax it at different times. The customers get to skip the capital outlay for the server plus get the compute they require, when they require it. This is the ideal cloud scenario.But what happens if a customer needs the performance of a whole box, most of the time? Then they're paying $1,400/month for $1,000's worth of computing. Or maybe, because they're reserving the whole box, they'll get a deal at $1,250/month by committing to a whole year. That deal is far less obviously good on both sides. It's basically a credit agreement at a 25% APR. Tread wisely!Enter serverless. It's the same financial mechanics as above, but you can slice the server far more thinly. Instead of renting out your one big server to seven customers at $200/month, you rent out individual function executions to 100 customers at $20/month. This now clears $1,000/month in profit instead of just $400/month. No wonder cloud providers love serverless!Again, if you only need a few functions executed every now and then, this works out for you as a customer too (at least in the short term). But if you execute enough functions to fill the computing power of a whole box, it's a terrible deal. Not just because you'll be paying more for the same clock cycles, but also because the lock-in is immense.The further down the rabbit hole you go with "cloud-native" services in serverless, the harder it'll be to climb out when you realize that you should own the donkey rather than rent it. And especially once you realize that paying to rent a whole donkey at the piece price of a hundred slices is an even worse deal than just renting the whole donkey by itself!Don't be fooled by serverless. There's no magic that can change the fundamental fact that if you need all the computing cycles of a computer, you ought to own that computer. And if you start off with a proprietary serverless setup, you might well find the lock-in impossible to escape by the time the rental math no longer works.The cloud is primarily for companies that have big swings in use – like Amazon's original AWS case of huge demand around Black Friday and Christmas, which left them with unused capacity for the rest of the year – or for early outfits that don't do enough business to either warrant owning a whole computer or spend so little on the cloud that it just doesn't matter. Serverless doesn't change that.

云计算爱好者将云计算的真正承诺放在无服务器功能和服务上,无服务器已经成为一个咒语,因为它仍然看起来很神奇,大多数人不会质疑基本原理。云计算和VPS都是基于古老的原则:批量购买,零售,如果客户对设备的需求不太大,可以节省成本。

Don't be fooled by serverless
相关推荐 去reddit讨论
  1. Google 发布「AI 全家桶」反击 GPT-4o !搜索引擎罕见大更新, 121 句「AI」道尽焦虑
    Google在Google I/O发布会上推出了多种新产品和升级,包括Gemini大模型、Gemma多模态大模型、AI in Google Workspa...
  2. 简单是优势,但为何复杂性还受欢迎?
    文章探讨了简单性和复杂性的优劣。复杂性在某些情况下更受欢迎,因为它传达了努力、精通和创新的信号。然而,简单性更易理解、使用、构建和维护,具有较低的运营成本...
  3. 一目十行:上下文快速阅读法
    基于上下文关系的阅读方法强调理解文本的整体语境,通过识别句子、段落甚至整篇文章中的逻辑关系和线索,来解读文本的含义。实践方法包括上下文预测、语境线索、重点...
  4. 太贴心!OpenAI发布了类人化重大版本:GPT-4o
    OpenAI发布了GPT-4o,具有实时翻译、情绪检测、语音指令理解、声音合成和图像理解等功能。GPT-4o将通过API免费提供给用户。OpenAI还发布...
  5. Python中读写Parquet文件的方法
    Apache Parquet是一种流行的列式存储格式,使用pyarrow包可以轻松读写Parquet文件。可以转换DataFrame为Parquet文件,...
  6. 使用 braft 构建应用,应该关注哪些指标?
    本文介绍了使用braft构建应用时需要关注的指标。braft是一个C++ raft框架,开发者可以基于其抽象接口实现自己的业务逻辑。文章从metrics入...
  7. Python读取NetCDF文件-裁剪&计算
    这篇文章介绍了使用xarray包处理NetCDF文件的方法。xarray是基于pandas的数据结构构建的,可以方便地处理多维数组数据。文章还介绍了如何根...
  8. How to Use Stable Diffusion Effectively
    稳定扩散是一个由多个组件和参数组成的流水线,它们共同工作产生输出。模型是流水线中最重要的组件之一,不同版本的模型会对输出产生不同影响。选择合适的模型、采样...
  9. sqlx: 一个优秀的rust异步SQL库
    Rust生态圈中的sqlx库是一个功能齐全的数据库访问和查询构建器库,支持多种数据库。它具有异步、编译时检查查询、与数据库无关、纯Rust等特点。sqlx...
  10. Python 潮流周刊#50:我最喜欢的 Python 3.13 新特性!
    本期刊共分享了12篇文章、11个开源项目和2则音视频,涵盖Python 3.13新特性、Asyncio工作原理、Python文件处理、数据科学错误、加速N...