README.markdown

Thin vs Puma

Run these two servers using the commands given in their source. Benchmark using:

ab -n 1000 -c 1000 http://localhost:8000/

I found thin can handle these requests without a problem. puma drops connections and causes ab to exit early.

With 200 threads, puma can handle 200 requests but does not respond promptly:

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:        4  915 1378.1      5    2991
Processing: 20096 20550 2063.8  20298   37213
Waiting:    20095 20487 2070.2  20238   37213
Total:      20160 21466 2654.2  20416   40203

Percentage of the requests served within a certain time (ms)
  50%  20416
  66%  20436
  75%  23104
  80%  23182
  90%  23262
  95%  23270
  98%  23270
  99%  40203
 100%  40203 (longest request)

async.ru

# thin start -R async.ru -p 8000
# rackup async.ru -s thin -p 8000
# rainbows async.ru -c rainbows.conf -p 8000

require 'eventmachine'

app = lambda do |env|
  cb = env['async.callback']
  EM.add_timer(20) do
    cb.call [200, {'Content-Type' => 'text/plain'}, ['Hello, world!']]
  end
  throw :async
end

run app

goliath.rb

# ruby goliath.rb -svp 8000

require 'goliath'

class HelloWorld < Goliath::API
  def response(env)
    cb = env['async.callback']
    EM.add_timer(20) do
      cb.call [200, {'Content-Type' => 'text/plain'}, ['Hello, world!']]
    end
    [-1, {}, []]
  end
end

rainbows.conf

Rainbows! { use :EventMachine }

sync.ru

# puma sync.ru -b tcp://0.0.0.0:8000 -t 0:1000

require 'eventmachine'
Thread.new { EM.run }

app = lambda do |env|
  response = nil
  EM.add_timer(20) do
    response = [200, {'Content-Type' => 'text/plain'}, ['Hello, world!']]
  end
  Thread.pass until response
  response
end

run app