lonex / em_vs_node_http Goto Github PK

- Background
  The purpose of the project is to compare the performance between Node.JS and EventMachine for the HTTP
  traffic.
  
  The plan is to test the following scenarios:

  - The server response against the fast http request.
  - The server response against the slow http request. E.g. slow network link or big file upload.

- Summary
  Other than the performance difference doing the sequential request (Test 1), the http server implemented in 
  Node and EventMachine are really close, namely Test {2,3,4} below.

             EM       Node
            =====     ====
    Test 1   20.9     4.3
    Test 2   2.4      2.4
    Test 3   2.5      2.5
    Test 4   3.8      4.1

- Configuration
 
  - The hardware
    - Server
      OS: Ubuntu 11.04
        Linux fermi 2.6.38-8-generic #42-Ubuntu SMP Mon Apr 11 03:31:24 UTC 2011 x86_64 x86_64 x86_64 GNU/Linux
      Processor: 1.66 GHz Intel Core 2 Duo
      Memory:    2.GB
 
    - Client
      OS: Mac OSX 10.6.7, 
        Darwin Wens-MacBook-Pro.local 10.7.3 Darwin Kernel Version 10.7.3: Sun Mar  6 13:37:56 PST 2011; 	
             root:xnu-1504.14.2~1/RELEASE_X86_64 x86_64
      Processor: 2 GHz Intel Core i7
      Memory:    8 GB 1067MHz DDR3

    - Network
      The client and the server communicate through a wireless LAN link 54Mbps.

  - The software
    - Client
      - Install httperf (0.9.0) on the client

      Before installation, depends on the system settings you have on your OSX installation, you may need to modify 
      /usr/include/sys/_structs.h to increase the FD_SETSIZE, e.g. from 1024 to 10240. 

      - Before running httperf as root, tune the system configuration temporarily.

      > sudo -i
      > sysctl -w kern.maxfiles=65536 kern.maxfilesperproc=32768
      > ulimit -n 12000

    - Server
      - Ruby version: ruby 1.9.2p180
      - Install node.JS (0.4.7) and eventmachine (0.12.10), eventmachine_httpserver(0.2.1) on the server.
      - The Node version uses the 'http' module coming with the node installation. The EventMachine version
        uses the http server from the eventmachine project, https://github.com/eventmachine/evma_httpserver.
        Since the HTTP server has overhead, e.g. html parsing, in addition to the underlying framework, the
        server implementation coming with Node and EventMachine was used.
        The server code is copied from the documentation of Node and EventMachine.

- Source code
  - EventMachine version: simple/em/server.rb
  - Node version: simple/node/server.js  

- Test result
  - Response time graph: simple/em_vs_node.pdf, which is based on the numbers listed in simple/em_vs_node.numbers.
  - The httperf benchmark results are under simple/em/perf, and simple/node/perf
  - sequentail_6000.txt (Test 1): this test includes the total of 6000 connections, each connection is created upon 
                         the tear-down of the previous connection, one at a time. One request is made within
                         each connection.
  - concurrent_2000_20.txt (Test 2): this test includes a total of 2000 connections, the ramp-up of the request is 
                         20 new connections per second. One request is made within each connection.
  - concurrent_2000_100.txt (Test 3): this test includes a total of 2000 connections, the ramp-up of the client is
                         100 new connections per second. One request is made within each connection.
  - concurrent_3000_200.txt (Test 4): this test includes a total of 3000 connections, the ramp-up of the client is 
                         200 new connections per second. One request is made within each connection.

- TODO
  - Finish the tests for the 2nd scenario, the slow client connection.