Introduction to the EEMBC BrowsingBench^® Benchmark

The BrowsingBench® benchmark provides a standardized, industry-accepted method of evaluating Web browser performance. It targets smartphones, netbooks, portable gaming devices, navigation devices, IP set-top boxes by measuring hte complete user-experience - from the click/touch on a URL to final page rendered and scrolled on the screen.

BrowsingBench benefits processor vendors, operating system and browser developers, and system developers. It helps determine effectiveness of associated hardware and software components when processing and displaying Web pages – both for performing the individual subtasks necessary to display a page, as well as the total composite time for processing the entire page. It allows a user to fairly and consistently compare browsing performance across different hardware platforms running a wide variety of software browser implementations.

General features:

• Targets both development platforms and finished commercial devices
• Local server support (WLAN and/or Ethernet) for stable and 100% repeatable results
• Support for any display size
• Utilizes web pages from multiple worldwide regions with a range of content and workloads
• Tests browser features that affect performance to ensure fair representation
• Supports power consumption and/or battery life measurement

BrowsingBench highlights the strengths and weaknesses of all system components involved in the browsing experience including the entire system, processors, operating systems, and browsers.

Theory of Operation

BrowsingBench runs a PHP site under an Apache server on a remote system. It can run from a USB stick, a local bootloader, a virtual machine, or a standalone installation. Once the server is running:

1. The user connects to the server via an IPv4 address using a browser on the device-under-test (DUT)
2. The user runs the benchmark on the DUT from within the browser
3. The server pushes hundreds of pages to the DUT (this may take 10-30 minutes)
4. The server measures the load-time of each page, as well as basic DOM rendering compliance
5. The server reports a score, which is a scaled geometric mean of the page load-time speed-up compared to a reference platform

At no time does the server fetch data from the WAN: the entire benchmark is self-contained between the client and server.

This simple capability can measure the impact from changing any part of the subsystem, including the internet connection speed or medium, the browser, the device under test CPU and RAM, etc. In addition, the number of runs can be changed to analyze battery life on hand-held devices.

The pages served by the benchmark were captured form the internet to represent a cross-section of typical usage. Multiple pages from each of the following sites were captured, including ads and javascript, which reflects actual replay of real-word traffic.

For BrowsingBench version 2, the following sites were surveyed:

• baidu.com (4 pages)
• cnet.com (3 pages)
• digg.com (1 page)
• eembc.org (2 pages)
• elektronik.de (1 page)
• facebook.com (1 page)
• google.com (1 page)
• nationalgeographic.com (3 pages)
• sina.com.cn (3 pages)
• uk.yahoo.com (4 pages)
• wikipedia.rg (2 pages)

Engineering Studies

BrowsingBench can also be configured a number of different ways, including a battery life test to measure the rundown of the DUT: