Many reports on web page sizes issued in recent years point to the same conclusion: the web has a weight problem. The web seems to be gaining weight each year despite the fact that study after study has shown a strongly negative reaction from users to heavy web pages and resulting loading times.
We decided to investigate this issue from a mobile web perspective. Picking up from where Guy Podjarny of Akamai left off, we examined the top 100 sites as ranked by Alexa.
Given our mobile focus we did things a little differently: rather than just requesting each page with a different browser window size (to trigger responsive design behaviours) we emulated a range of different devices.
Note: We use the phrase dynamic range to describe the extent to which a site is capable of producing lighter views of the same content depending on the nature of the visiting device. The dynamic range of a page is the ratio of the heaviest version of the page to its lightest. Normally such behaviour is triggered based on device characteristics but network connectivity is increasingly being utilised also. A smartphone at the end of a GPRS connection or poor airport WiFi is effectively a low-end phone.
The results are dramatic. Here is the dynamic range of the top 10 sites:
Site | Dynamic Range | Smallest Size (KB) |
Largest Size (KB) |
|
---|---|---|---|---|
google.com | 40 | 8 | 305 | |
facebook.com | 34 | 7 | 241 | |
youtube.com | 21 | 24 | 517 | |
yahoo.com | 13 | 34 | 444 | |
baidu.com | 9 | 8 | 68 | |
wikipedia.org | 1 | 100 | 100 | |
live.com | 55 | 5 | 271 | |
qq.com | 5 | 478 | 2,271 | |
amazon.com | 154 | 14 | 2,131 | |
taobao.com | 37 | 40 | 1474 |
The main takeaway point is that, despite the general trend towards heavier pages, the top web brands excel at delivering light pages when necessary. The average dynamic range across all sites measured was 26, with a median of 15. In other words, these web brands were capable of changing weight of their primary entry point by a factor of 26, on average. Of all one hundred sites tested only four sites made no attempt to cater for different devices by lightening content, and one of these (Wikipedia) does it for most of the content pages on the site even though it doesn’t do so on the home page. In general, the more global the brand the better the site’s dynamic range.
So what does this mean? It means that if you’re a global web brand you can’t afford to limit your audience with a web presence that excludes devices and territories. Visually squishy layouts are not sufficient and the top sites know this; they cater to the lumpy, uneven web, the realistic web of mixed devices, poor bandwidth and mobile-only nations.
As an example of the importance of this it is recommended to read a very interesting post from Chris Zacharius who talks of his experiences in removing weight from YouTube pages. They found that lightening the default YouTube “watch page” literally opened up entire new markets for them.
The ensuing conversation on Hacker News is also worth reading, particularly this comment.
Conclusion
To paraphrase socialite Wallis Simpson, on the web you can never be too light or too fast. The global web is lumpy and uneven, consumed on diverse devices with myriad connectivity rates in widely varying conditions. It is a mistake not to learn from the top brands—one size decisively does not fit all.
Notes & Methodology
Some of the dynamic range figures shown are not truly indicative of what is happening. Two notable examples are xhamster.com (NSFW) and sina.com.cn. In both cases the very high dynamic range figure is due to a placeholder page offering the user a choice of onward destinations.
By contrast, the Wikipedia figure of 1.0 under-represents the reality. In fact, all of the article pages on wikipedia.org (the vast majority of the site) have a dynamic range of about 2 or so.
I used the excellent webpagetest.org service to run these tests. I drove this tool using their REST API. A script was used to configure the host browser’s window size and user agent string before fetching each page. Our script is as follows:
1 2 3 |
setViewportSize %s %s setUserAgent %s navigate %s |
In each case the main entry point for the site was fetched and the browser was allowed to follow redirects.
The user agents used will be familiar to anyone who has used Prism:
- iPhone: Mozilla/5.0 (iPhone; CPU iPhone OS 5_0_1 like Mac OS X) AppleWebKit/534.46 (KHTML, like Gecko) Version/5.1 Mobile/9A405 Safari/7534.48.3
- Nokia 6300: Nokia6300/2.0 (04.20) Profile/MIDP-2.0 Configuration/CLDC-1.1
- Blackberry 8520: BlackBerry8520/4.6.1.259 Profile/MIDP-2.0 Configuration/CLDC-1.1 VendorID/1
- Chrome: Mozilla/5.0 (Macintosh; Intel Mac OS X 10_8_2) AppleWebKit/537.22 (KHTML, like Gecko) Chrome/25.0.1364.99 Safari/537.22
These UAs were chosen since they represent a wide range of device types in common use, from a low-end Nokia to an iPhone and hence tend to elicit a wide range of behaviours from websites.
For transparency I’ve attached an Excel file containing all of the test data to the end of this post.
Leave a Reply