Study of the performance of internet browsers (Part 2)
	by William Rang, CTO

  The purpose of the preceding study was to show that the recent changes to the new versions of internet browsers, namely, the number of simultaneous connections (sockets) open to a single server, can improve the performance of page loading.

The aim of the present study is to confirm whether these technological advances (number of simultaneous connections; new, more efficient JavaScript machines) offer a real gain in performance for the end user.

Results of the study

To collect a reliable statistical sample, 2500 measurements were made on each browser. A panel of five websites were used as targets. A detailed methodology is provided in section 2.

Overall loading performance

The graph below shows the average overall performance by browser:

Remarks:

• Google Chrome is the most efficient navigator tested.
• A significant change was noted between the three latest version of Internet Explorer. IE8 beta 1 thus slightly outperforms Firefox versions 3.0 and 3.1 beta 1.
• Firefox versions 3.0 and 3.1 beta 1 show essentially identical performances.
• The strategy of increasing the number of simultaneous connections to a single server is apparently a fruitful one, even if the design of the sites with, for instance, JavaScript processing locally on the client workstation, can considerably deteriorate performance levels (cf. example below).

Performance details by target site

  Google

  PagesJaunes

TF1

ViaMichelin

  01net

Methodology of the study

The methodology followed in this study takes account of the following:

Technical scope

Measurements were carried out on a dual-core 2.4 GHz server with 1 GB RAM, running Windows XP.
The Internet browsers monitored:

• Google Chrome
• Firefox 3.0
• Firefox 3.1 beta 1 (TraceMonkey JavaScript machine enabled)
• Internet Explorer 6
• Internet Explorer 7
• Internet Explorer 8 beta1

The capability for opening simultaneous connections to a single HTTP 1.1 server

 :

The sites tested: 

• www.google.fr
• www.pagesjaunes.fr
• www.tf1.fr
• www.viamichelin.fr
• www.01net.fr
  

Below, the volume of data downloaded or uploaded for each site (*) :

(*) The volume may vary for some pages depending on the dynamic objects present (eg. advertising).
Number of requests needed (i.e. number of objects) to load the entire content:

Particular points:
The browser’s cache was emptied prior to each measurement.
The loading time measured began when the navigation was initiated and ran until the end of the browser’s activity (no more object to load). A different method was applied for the site Via Michelin because of the number of asynchronous exchanges on the site. We set the moment the French flag appears on the page as the end of loading for this site.

Geographical scope

Measurements were carried out using a Free ADSL connection with a download rate of 5 MB/second.

Sample size

The total number of measurements over the period from 30/10/2008 to 7/11/2008: