Distributed proxies for browsing privacy: a simulation of flocks

• Authors: Martin S. Olivier
• Venue: SAICSIT '05 Proceedings of the 2005 annual research conference of the South African institute of computer scientists and information technologists on IT research in developing countries
• Year: 2005

In previous work we introduced an anonymising proxy scheme --- called Flocks --- to be used for browsing privacy. Flocks is similar to Crowds in that each proxy randomly decides whether to forward any request it receives to the destination Web server, or whether to forward the request to another proxy. In this manner request chains are formed that hide the details of an originator of a request from the destination server as well as from the various proxies.Flocks differs from Crowds (and similar Privacy-enhancing Technologies or PETs) because it caches pages that are requested. Unlike related PETs, Flocks is intended for deployment within an organisation. Caching minimises the need for communication between an organisation and external content providers, decreasing cost and potentially increasing access speed. Logging in Flocks is designed to balance privacy with the need to conduct forensic investigations when required (with safeguards to prevent unauthorised breaches of privacy).Two parameters determine the behaviour of Flocks: α is the probability with which any proxy will forward a request to an external server (rather than to another proxy) and N is the number of proxies in the system. In previous work we analytically determined the impact of these two parameters on privacy and performance aspects of Flocks.The current paper reports on simulations that were performed to gain deeper insight into the behaviour of Flocks. The simulations confirm the analytic results of our previous work. They also shed light on performance-related issues such as the number and positions of access to external servers, saturation levels and traffic patterns. This information will be useful to decide on appropriate values of α and N from a performance point of view.The simulations also highlight the problem of overly long chains that will occasionally occur. A simple solution is proposed and tested empirically. The privacy and performance implications of this solution are discussed; it is found that the solution is usable, but has a profound impact on the choice of the a and N parameters.