Topics: Big data in networks; Traffic characterization and traffic models for networks
Authors: Charles Thomas, Jeff Kline and Paul Barford (ComScore, USA)
Presenter bio: Charles Thomas currently lives in Madison, Wisconsin, USA. He spent 26 years at the University of Wisconsin, Madison, where he was developed novel research computing and large-scale network management softwar systems. Charles designed and built the high-performance JavaScript telemetry tag that was central to data collection for the fraud detectio company, MdotLabs. In 2015, MdotLabs was acquired by comScore, a leadin web metrics company, where he now works. Currently he is engaged in research and software development in the AdTech problem space.
Abstract:
Collecting information from web clients
without explicit input from users is important in a variety of contexts
including content customization, experience personalization, accounting
and online advertising. A standard approach for gathering web client
telemetry is through deployment of Javascript instrumentation that is
placed either directly on web pages or through third-party "tags" that
are referenced in web pages. In this paper we present a design study of
web client measurement methods. The objective of our work is to enhance
understanding of best practices in web client measurement toward the
goal of developing future tags that are reliable, robust and efficient.
We begin by conducting a detailed examination of Javascript
instrumentation collected from five well known third party services. Our
analysis shows that these code-bases have diverse capabilities and
return a broad range of client characteristics. Next, we describe a web
client measurement framework and an implementation that we call
IntegraTag, which enables us to examine details of performance, accuracy
and reliability through live deployments. We use IntegraTag to conduct
case studies of tag behavior on a single website which resulted in over
500K page-loads, and on a publisher network which resulted in over 150M
page-loads. We report on the wide range of client characteristics
returned by IntegraTag, as well as it's performance and robustness.
