Authors: Kyle Vogt <[email protected]> and Emmett Shear <[email protected]>

Copyright (c) 2008, Justin.tv, Inc."

License: MIT

Twice was designed to solve some of the most challenging problems faced while trying to scale a web application. It is written in Python using the Twisted networking framework. Twisted uses a single-threaded, asynchronous model, so it's good at handling high concurrency situations.

• In-memory cache
• Basic templating language to render dynmaic page elements
• Intelligent reverse proxy (Squid like functionality)
• HTTP header and cookie inspection
• Written in Python using the Twisted event framework

Twice is a reverse-proxying webserver with an in-memory cache and basic templating language. It serves each incoming request using its cache when possible. Otherwise, requests are passed on to the backend servers. The response is then cached for a finite amount of time. When the cache expires, it will be refreshed lazily when a new request for that particular page arrives.

You can configure Twice on a per page basis by adding extra HTTP headers to your application server's responses.

It is best to run Twice behind a hardened web server (like Apache, Lighttpd, or Nginx). These programs are better suited to serve static content like images and css files. Other requests hit Twice, which serves cached responses when it can. If Twice can't serve a page, it will proxy the request through to the application server(s) and intelligently cache the response.

At Justin.tv we use Twice to serve the majority of dynamic page views (up to 25 million pages per day). Here are some of the things it does:

• Temporarily caches pages that update frequently
• Renders cached pages in foreign languages by inspecting HTTP headers
• Renders usernames at the top of every page by inspecting browser cookies
• Renders page view counts into cached pages on demand by reading memcached keys

Twice cut our peak application server load by 78% and reduced our average page load time by 50% overnight.