Chapter 14. Case Study: Ad Delivery
The process of delivering advertisements on the Internet constitutes a possible application of feedback principles. It exhibits many of the characteristics that make feedback mechanisms desirable:
The laws governing the process are either not known or only incompletely known; the process itself is opaque.
The governing laws may change silently over time.
The process is subject to random changes (because of fluctuations in web traffic).
Yet, there is a clear goal; typically, this is the number of ad impressions to show every day or the amount of money to spend.
What is desired is a reliable mechanism for executing the plan in the face of uncertainty about the process and subject to the random fluctuations in web traffic.
The Situation
The specific situation we will consider consists of a publisher or advertising network that displays ads—ours as well as those from competing advertisers. We cannot directly control how often our ads are shown. Instead, we can name the maximum price that each showing (or impression) is worth to us, after which the publisher will make a selection based on this offer. All we know is that a higher offered price tends to result in a greater number of ad impressions.
The system we want to control is the publisher. The control input is the price, and the control output is the number of impressions served. Knowing that a higher price means more impressions provides the tiny but necessary bit of process knowledge required to set ...
Get Feedback Control for Computer Systems now with the O’Reilly learning platform.
O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.