In Search of: Reliable Usage Data on the World Wide Web
The WWW is currently the hottest testbed for future interactive digital systems. While much is understood technically about how the WWW functions, substantially less is known about how this technology is used collectively and on an individual basis. This disparity of knowledge exists largely as a direct consequence of the decentralized nature of Web. Since each user of the Web is not uniquely identifiable across the system and the system employs various levels of caching, measurement of actual usage is problematic. This paper establishes terminology to frame the problem of reliably determining usage of WWW resources while reviewing current practice and their shortcomings. A review of the various metrics and analyses that can be performed to determine usage is then presented. This is followed by a discussion of the strengths and weaknesses of the hit-metering proposal [Mogul and Leach 1997] currently in consideration by the HTTP working group. Lastly, new proposals, based upon server-side sampling are introduced and assessed against the other proposal. It is argued that server-side sampling provides more reliable and useful usage data while requiring no change to the current HTTP protocol and enhancing user privacy.