Average Web Page Size Triples Since 2003
The size of the average web page has more than tripled since 2003. From 2003 to 2008 the average web page grew from 93.7K to over 312K (see Figure 1), some 233% (Domenech et al. 2007, Flinn & Betcher 2008). During the same five-year period, the number of objects in the average web page nearly doubled from 25.7 to 49.9 objects per page. Longer term statistics show that since 1995 the size of the average web page has increased by 22 times, and the number of objects per page has grown by 21.7 times.
A web page is made up of a container object (CO) and external objects (EOs). The CO is usually an XHTML file that references EOs such as images, audio, video, and external CSS and JavaScript files. Most non-textual EOs are usually pre-compressed, so you'll only see the benefits of HTTP compression on XHTML and on some CSS and JavaScript files. For more than 60% of web pages, the CO occupies less than 50% of total page size. The average CO takes up about 44% of total page size (Yuan, Li, & Chi 2005). Therefore, no matter how efficient your XHTML optimization and HTTP compression, the greatest improvement in web page performance that you can expect from XHTML optimization will be less than 50%. For a 300K home page, on average, you'll still need to download at least 150K of EOs, even after optimizing and compressing your textual data. You can see why it is important to optimize your entire web page, including multimedia, in order to make significant performance gains. Over the past calendar year, the thousand most popular home pages grew by 24.2% from December 2006 to December 2007, that is, from 250K to 310.4K (see Figure 2). At this growth rate the average home page will exceed 385K by the end of 2008. The number of objects grew by 14.5% in 2007 from 44.2 in December 2006 to 50.6 objects per page in December 2007. An October 2007 survey of the top 50 retailers revealed that the average number of objects per page was 60.5 (Flinn & Betcher 2007). With the average number of objects per page above 50, the latency due to object overhead now dominates most web page delays. From 2003 to 2008 web page size has more than tripled and the number of external objects has nearly doubled. So narrowband users (56K and ISDN) have experienced slower response times over time. Conversely, broadband users have experienced somewhat faster response times over time. For broadband users the average download time of the Keynote Business 40 Internet Performance Index (KB40) has decreased from 2.8 to 2.33 seconds from Feb. 2006 to Feb. 2008 (see Figure 3 and Berkowitz & Gonzalez 2008). So the increase in the average speed of broadband has more than kept pace with the increase in the size and complexity of the average web page. That is one reason why broadband users expect faster response times. Yet narrowband users have experienced slower response times as web page size has increased. In a 2006 survey of over 21,500 non-framed web pages, Ryan Levering and Michal Cutler found that the average web page contained 474 words, 281 HTML tags, and 41 links, 10 of which pointed outside the domain (Levering & Cutler 2006). They found that the average web page was 1440 pixels in height, about twice the size of the screen height. When the average web page is opened, more of what the user sees is graphics, not text. Graphics make up most of the objects in the average web page. Graphics cause most of the delays that users experience from object overhead. A subsequent survey in 2007 run for the author revealed changes in the composition of the average web page. Despite the widespread adoption of CSS, the 2007 survey found that 62.6% of web pages still use tables for layout and 32.8% use the font tag for inline style (Levering 2007). In an encouraging trend, the average table depth decreased by half from the 2006 survey, from 2.95 to 1.47. Complex nested tables can cause rendering delays with browsers because tangled tables must be parsed and rendered before your content displays. Between the 2006 and 2007 surveys, the average number of HTML elements per web page more than doubled from 281 to 592.6 elements per page. In the 2007 survey, 84.8% of web pages used the script element. The average size of external scripts was 8,845 bytes uncompressed, and 6,302 bytes compressed. Total script size was 68,812 bytes uncompressed and 49,738 bytes compressed. The average number of external scripts was 7, where 6 of which were unique. In the 2007 survey, 82.4% used the link tag, and 54.5% used the style tag (with an average of 2.27 style tags used internally). The average size of external style sheets was 6,575 bytes, and 4,457 bytes compressed. The total average style size was 15,175 bytes uncompressed, and 10,347 bytes compressed. Images were used in 91.6% of web pages in the 2007 survey. GIFs were used in 84.6% of web pages, up from 77.9% in 2006 (see Table 1). JPEGs were used in 64.5% of web pages, up from 55.8% in 2006. PNGs were used in 32.2% of web pages, up significantly from 7.2% in 2006. BMPs were used by 0.8% of web pages in 2006. BMPs in particular can be quite large as they are usually not compressed. They should be avoided for Web use. If the images in the average page were grouped together, the image area would be 49,144 pixels, or an image of 221 by 221 pixels in dimension (see Figure 4). A large-scale study confirmed these findings with over 75% of HTTP requests resulting from images (Gonzalez-Canete, Casilari, & Trivino-Cabrera 2007). The use of streaming media on the Web has increased by more than 100% each year (Li et al. 2005). From 2000 to 2005 the total volume of streaming media files stored on the Web grew by more than 600%. More than 87% of all streaming media is abandoned by users in the first 10 seconds, however, wasting up to 20% of server bandwidth (Guo et al. 2005). While only 3% of server responses are for videos, they account for over 98.6% of the bytes transferred (Gill et al. 2007). In true Pareto fashion, about 10% of the most popular videos on YouTube account for nearly 80% of the views, making caching an appealing performance enhancement (Cha et al. 2007). Overall, for videos longer than 30 seconds, about 13% of home and 40% of business users experience quality degradation with their streaming media, caused by re-buffering, stream switching, and video cancellation. For sessions longer than 300 seconds, the results are even worse. As broadband penetration has increased, videos have grown in size, bit rate, and duration (see Figure 5). In 1997, 90% of videos were under 45 seconds in length (Acharya & Smith 1998). In 2005, the median video was about 120 seconds long (Li et al. 2005). By 2007, the median video was 192.6 seconds in duration (Gill et al. 2007). The median bit rate of web videos grew from 200Kbps in 2005 to 328Kbps on YouTube in 2007. So by late 2007, the median video weighed in at over 63MB in file size. On YouTube, the average video size is 10MB, with over 65,000 new videos added every day. The purpose of this article is to show trends in the size and composition of the average web page. Some readers have asked why this is happening. Web 2.0 technologies such as Ajax certainly contribute to the increase in the number of objects per page, as well as the growth in JavaScript file size. Dynamically generated sites from content management systems are typically not as optimized as hand-tuned sites, and often carry over site-wide CSS, JavaScript, and page components to every page on a site. As broadband becomes more widespread web designers have created more elaborate designs, assuming that a large proportion of their audience is on broadband, or ignoring dial-up users entirely. At the time of this writing, however, 43% of US households are still on dial-up, according to our latest bandwidth report. Some solutions are as follows: Within the last five years, the size of the average web page has more than tripled, and the number of external objects has nearly doubled. While broadband users have experienced faster load times, dial-up users have been left behind. With the average web page sporting more than 50 external objects, object overhead now dominates most web page delays. Minimizing HTTP requests by using CSS sprites, combining JavaScript or CSS files, reducing the number of EOs, and converting graphic effects to CSS while still retaining attractiveness, has become the most important skill set for web performance optimizers.
Figure 1: Growth of the Average Web Page
Anatomy of a Web Page
Growth of the Top 1000 Web Pages in 2007
Figure 2: Growth of the Average Top 1000 Home Page
Response Time Trends - The Bandwidth Divide
Figure 3: Average KB40 Web Site Performance over Broadband from Feb. 2006 to Feb. 2008 (Source: Keynote Systems, Inc.)
Average Web Page Characteristics
Change in the Average Web Page over Time
Use of JavaScript in the Average Web Page
Use of CSS in the Average Web Page
Use of Images in the Average Web Page
Image Encoding Frequency 2006 Frequency 2007 GIF 77.9% 84.6% JPEG 55.8% 64.5% PNG 7.2% 32.2% BMP 0.8% - Table 1: Image Usage in the Average Web Page (2006 and 2007 surveys)
Figure 4: Average Aggregated Graphic Area in the Average Web Page
The Growth of Multimedia on the Web
Figure 5: Growth in Duration of Web Videos
Why the increase in the Average Web Page?
Conclusion
Further Reading
By website optimization on 28 Apr 2008 AM
Comments
All,
Comments should now work, they were inadvertently turned off after our upgrade to MT 4.X.
By: website optimization on May 2, 2008 9:46 AM
I've added a section on the reasons why the average web page has grown in size and complexity per reader request.
By: website optimization on May 2, 2008 9:47 AM
Great info, but I was confused re: "At the time of this writing, however, 43% of US households are still on dial-up, according to our latest bandwidth report."
When I read the stats at http://www.websiteoptimization.com/bw/0804/ I didn't get that 43% of US households are still on dial-up...I read it as in ALL US households, broadband penetration was at 57% and then in households that actually have an internet connection, 88% were using broadband (meaning less than 12% of homes w/ an internet connection were using dialup). So of that 43% not on broadband, some are on dialup and some have NO internet connection.
Can you clarify which it really is? 43% of US households still being on dialup seems high in comparison to what I've read elsewhere.
By: Megan on May 28, 2008 12:26 PM
Very interesting and detailed post. I included it in my International Marketer Review Blog Carnival that was posted today.
I look forward to reading your new book.
By: Cindy King on August 9, 2008 5:39 AM
Can you show an example of what progressive enhancement looks like? Is there a site out there doing this now? The only way I can think to know if users are using dial-up or broadband is include additional scripting to measure the speed of their downloads, adding to the already burgeoning load. As well some functionality just can not be replicated with less which defeats the purpose of some web content/applications. Certainly there is a lot of gratuitous media and graphics, but in increasing number of cases that is the content.
I think most folks who do use dial-up at home (I also echo the question of Megan on the validity of 43% on dial-up), probably do so because of the lack of broadband availability not because of choice or limited income. So an alternative solution is to kick the butts of providers to get that access. The US is falling behind other countries in speed of connectivity.
By: jim on August 19, 2008 9:23 AM