Influences of Node
Benevolent Dictator Emeritus

July 2, 2012

Node was a long progression of attempts to build websites. It was influenced by a lot of software: CouchDB, Ebb, Flow, FUSE, libebb, libeio, libev, libircclient, Merb, Mongrel, NGINX, nginx-ey-balancer, Ragel, Ruby on Rails, SqueezeBox, timber lang, XUpload And probably others. This is a story of what I was doing before Node was first released in June 2009.

2004-11-10. Rochester, New York.

http://tinyclouds.org/topology.pdf

2006-08-20. Valparaiso, Chile.

2006-08-10. Valparaiso, Chile. Each night I’ve been working until the cold winter morning in order to discover a method of communicating via clouds. As of yet, I am unsuccessful.

2006-08-15. Valparaiso, Chile. One project is almost done and I can put it online soon. It’s a theorem database and automated proof thingy for topological spaces. I made it so that I could learn Ruby on Rails - which is hot.

2006-12-09. Buenos Aires. I just got a contract to do a big website. $ + work = travels to tierra del fuego in January. 2006-12-18. Buenos Aires. HTTP upload progress bars are rather obfuscated- they typically involve a process running on the server keeping track of the size of the tempfile that the HTTP server is writing to, then on the client side an AJAX call is made every couple seconds to the server during the upload to ask for the progress of the upload. This is pretty ridiculous. http://four.livejournal.com/730831.html Ebb web server. Like Thin but using libev instead of EventMachine

http://ebb.rubyforge.org/

http://squeezebox.rubyforge.org/

Precursor to libuv–the just the C parts of Ebb. Unix only.

http://tinyclouds.org/libebb

2007-12-06. Cologne. I’m working on a new HTML templating language using javascript. The language uses indention white space to build a XML hierarchy, the Haml syntax. In and around the XML one may use javascript for loops, conditionals, and subrutines. The attributes of a XML node are javascript objects. [...] I’ve written a proof of concept in Ruby using Ragel and SpiderMonkey. [...] It will allow front-end developers to more work independently of server-side programmers.

http://four.livejournal.com/820601.html

EngineYard contracted me to make NGINX load balance in a similiar way to haproxy. Before I began I demonstrated the need for this: 2009-01-26. Cologne. The result is what every system admin knows intuitively: Rails gets worse over time if loaded down with connections. The culprit is most likely MRI’s crappy thread implementation. While Rails serves 1 request, 29 threads are there sitting idle sucking up resources. http://four.livejournal.com/955817.html

2009-01-30. Cologne. i’m making a file system with only one thread! 2009-02-19. Cologne. i’m listening an album which is stored on the file system i wrote. EngineYard had contracted me to build a FUSE file system on top of a nascent (and now dead) DynamoDB clone. It was killed after only a few weeks. However the origin of the JavaScript “comma-first style” began with how I styled the C in this project. http://four.livejournal.com/964924.html

2009-02-09. Cologne. I’m going to write a special thin web server tied to the V8 javascript interpreter (Mongrel for V8 if you will) write a special evented TCP library for V8 (probably by wrapping liboi). Does this exist in public already? Certainly google has such a thing bundle them with a HTTP API for modifying server-side documents [...] I think this design will be extremely efficient and support very high loads. http://four.livejournal.com/963421.html

Ryan (painfully hungover) and Tim at Curucamp 2009 in Vienna.

Date: Sun, 1 Mar 2009 02:46:33 +0100 Subject: solution From: Ryan Dahl To: Tim Becker okay. going to do it in javascript. main thing is tcp. all will follow from there. maybe we’ll just use databases instead of file systems. (we can cache image data inside javascript if need be) So. TCP API:

TCP.connect(host , port , onConnect , onRead , onDrain , onClose);

Example: function myOnConnect () { this.write(’GET /index.html HTTP/1.1’); this.write(’\r\n\r\n’); } TCP.connect("google.com" , 80 , myOnConnect); I will implement DOM’s timer thing setTimeout() so that it uses libev timers. thoughts? i think this is extremely simple and will solve the world’s problems.

Date: Sun, 1 Mar 2009 09:40:29 +0100 Subject: Re: solution To: Ryan Dahl From: Tim Becker It is extremely simple, feels like it’s missing something, but I can’t think of what at the moment apart from ‘close‘. Anyway need to pack, see you! -tim

Date: Sun, 1 Mar 2009 14:49:06 +0100 Subject: Re: solution From: Ryan Dahl To: Tim Becker So. The idea is shifting in my head. Instead of presenting this primarily as a web server with some extension libraries for TCP lets present it as a set of evented libraries. Initially we will provide TCP.connect() // described above HTTP.serve(’localhost’, 80 , onRequest)

But later we’ll provide TCP.serve, HTTP.connect, MySql.connect ? This will be the framework that we eventually build our HTTP-only web server (where you upload javascript to change the behavior through HTTP). But we’ll release this set of libraries initially. Maybe call it netv8? ./netv8 myscript.js Would be nice to have some way of loading external js files (perhaps through TCP?)

Date: Sun, 1 Mar 2009 14:58:05 +0100 Subject: Re: solution From: Ryan Dahl To: Tim Becker > Maybe call it netv8? > ./netv8 myscript.js better name: "node" ?

./node irc_client.js

WHEREAS, The usage of threads has complicated computer programming; and WHEREAS, V8 javascript comes free of I/O and threads; and WHEREAS, Most operating systems do not provide asynchonous file system access. Now, therefore: This set server and client libraries were made to build simple but fast servers. They are provided free of charge under a permissive simple license. Submitted by Ryah Dahl, Tim Becker, March 1, 2009

Initial Announcemnet: 2009-05-31. Cologne. Node is a new server-side javascript project. It provides a purely event-based interface to I/O: TCP server and client Standard setTimeout() setInterval() timers Asynchronous file I/O HTTP server and client Node’s main focus is on performance and efficiency.

http://four.livejournal.com/1003191.html