Celluloid::IO-based web server
Switch branches/tags
Nothing to show
Pull request Compare This branch is 554 commits behind celluloid:master.
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
benchmarks
bin
examples
lib
spec
tasks
.gitignore
.rspec
.travis.yml
Gemfile
LICENSE.txt
README.md
Rakefile
logo.png
reel.gemspec

README.md

Reel

Build Status

Reel is a fast, non-blocking "evented" web server built on http_parser.rb and Celluloid::IO. It's probably most similar to Goliath, but thanks to Celluloid can be easily composed with multithreaded applications.

Connections to Reel can be either non-blocking and handled entirely within the Reel::Server thread, or the same connections can be dispatched to worker threads where they will perform ordinary blocking IO. Reel provides no built-in thread pool, however you can build one yourself using Celluloid::Pool, or because Celluloid already pools threads to begin with, you can simply use an actor per connection.

This gives you the best of both worlds: non-blocking I/O for when you're primarily I/O bound, and threads for where you're compute bound.

Is It Good?

Yes, but it has room for improvement. A "hello world" web server benchmark, run on a 2GHz i7 (OS X 10.7.3). All servers used in a single-threaded mode.

Reel performance on various Ruby VMs:

# httperf --num-conns=50 --num-calls=1000

Ruby Version        Throughput    Latency
------------        ----------    -------
Ruby 1.9.3          5263 reqs/s   (0.2 ms/req)
JRuby 1.6.7         4303 reqs/s   (0.2 ms/req)
JRuby HEAD          5650 reqs/s   (0.2 ms/req)
rbx HEAD            2288 reqs/s   (0.4 ms/req)

Comparison with other web servers:

Web Server          Throughput    Latency
----------          ----------    -------
Goliath (0.9.4)     2058 reqs/s   (0.5 ms/req)
Thin    (1.2.11)    7502 reqs/s   (0.1 ms/req)
Node.js (0.6.5)     11735 reqs/s  (0.1 ms/req)

All Ruby benchmarks done on Ruby 1.9.3. Latencies given are average-per-request and are not amortized across all concurrent requests.

Usage

Reel provides an extremely simple API:

require 'reel'

Reel::Server.supervise("0.0.0.0", 3000) do |connection|
  request = connection.request
  puts "Client requested: #{request.method} #{request.url}"
  connection.respond :ok, "hello, world"
end

Status

Reel is mostly in the proof of concept stage, working well enough you can point your favorite benchmarking utility at it, but not exactly speaking valid HTTP yet.

Contributing

  • Fork this repository on github
  • Make your changes and send me a pull request
  • If I like them I'll merge them
  • If I've accepted a patch, feel free to ask for commit access

License

Copyright (c) 2012 Tony Arcieri. Distributed under the MIT License. See LICENSE.txt for further details.