This report presents an analysis of the performance of two hundred Internet sites over a one month period November 2000. The sites examined range from major Internet destination sites to government agencies, from consulting companies to educational institutions. A similar study was performed in 1999 and can be found at http://www.e-insights.com/.

Noteworthy changes from 1999 include the substantial increase in the average page size (up from 73KB to 98KB), while the average page download time remained essentially constant at 4.3 seconds. Based on the stable page download time and the large change in page size, the performance of the systems and networks used to deliver these pages has shown marked improvement. Additionally, the distribution of page download time has reduced variance – meaning more predictable and consistent times. In substantial part, these improvements appear to be related to the increasing use of content distribution services such as Digital Island’s Footprint and Akamai’s FreeFlow. In order to better understand the impact of such services, E-insights is conducting research on this topic and expects to publish a report in February 2000.

On a negative note, the average availability of sites dipped significantly down to approximately 95% from 1999’s 97%. A part, but not all, of this decrease can be attributed to observed bi-modal behavior on financial services sites. Normally these sites are very fast, however at times of heavy load their response times become so slow as to be categorized as unavailable.

Also noteworthy was the increase in overall page complexity, in terms of the number of URLs per page, and the usage of scripts and frames, while at the same time overall quality increased. Interestingly, the usage of APPLETS decreased.

This year we have also added a section on page coding efficiency, i.e. calculating what percentage of the data actually sent from a server to a browser is in fact required to render the requested page. While the overall efficiency is relatively high, there are some site with very low coding efficiency, resulting in both lower response time to users, and presumably higher bandwidth charges to site operators.

________________________________________________________________________

The figures below show site performance by industry category and by server software, organized into Speed/Availability comparisons and Quality/Complexity comparisons. Speed is based on average download time, availability is based on the percent site availability, quality is based on the percent of pages which specify resources which cannot be retrieved, and complexity on the average number of URLs specified in a page.

Figure 1: Speed/Availability comparisons.

Industry Category Key –

While higher Quality and Availability are always better, there is a tradeoff between complexity and download time. More complex pages allow for a richer end user experience, but also in general are larger and take longer to download.

Figure 2: Complexity/Quality comparisons.

In examining Figures 1&2, the four categories that are in the quadrant denoting above average speed and availability, are also all in the above average quality and below average complexity quadrant. In comparison with 1999, overall quality has improved markedly. In addition, if the EDU category is excluded, all categories and Server Software vendors are grouped together much more tightly than in 1999, indicating more consistent and improved process control in general, which one would expect to see as the organizations and infrastructures mature.