Metrics Project: Measuring the Tor Network

The Metrics Project aims at measuring various aspects of the Tor network, including both contributed resources and network usage. These information are important, e.g., for deciding what parts of the network require more support to better meet user needs.

There are at least three aspects of the Tor network which shall be considered in this project:

• Relays: There are already statistics covering the data that relays report about themselves and which are confirmed in parts by the directory authorities. The TorStatus project evaluates the hourly published directories containing lists of all running relays and keeps track of their uptimes and advertised bandwidths. However, an aggregate view of the development of relays over an extended period of time is still missing. In addition to that, all information about provided bandwidth capacity as reported by the relays is not necessarily trustworthy. There has not been a comprehensive measurement of relay performance so far to validate bandwidth claims and relay stability.
• Bridges: A special type of relays is running as bridges. Bridges are relaying traffic, too, but are not listed in the public directory. The purpose of bridges is to allow censored clients to connect to the Tor network without their censor being able to prohibit access easily. The number and capacity of bridges is therefore not publicly known and has not been evaluated, yet.
• Clients: Little is known about usage of the Tor network. In particular, the number and location of clients are largely unknown. A rough estimation is that there are a few hundreds of thousands of clients in the network; a number which is very hard to verify. There are no recent studies on the number of Tor users which would help obtaining a better picture on usage of the Tor network. Clearly, all investigations of clients need to be performed with special care and in highly aggregated form to protect users' anonymity.

The Metrics Project is planned to run for three years from September 2008 to August 2011 (see section 5.7 of the three-year development roadmap for more details). Milestones are scheduled for every six months in this period. The following sections (yet only one) show progress made in these milestones.

February 2009

The first milestone of the Metrics Project is intended to evaluate public data about the Tor network, like network statuses and router descriptors. These evaluations shall help decide what data will be needed in the future to measure contribution to and usage of the Tor network. The following results are already available. Feedback is very welcome!

• Evaluation of Relays from Public Directory Data (PDF, 558K, last updated on June 22, 2009).
• Analysis of Bridge Usage in Tor (PDF, 76K, last updated on June 22, 2009).
• Evaluation of Client Requests to the Directories to determine total numbers and countries of users (PDF, 207K, last updated on June 25, 2009).
• All scripts and sources used in these evaluations (not the data, unfortunately) are available in a public Git repository that can be cloned using the following command:
  git clone git://git.torproject.org/git/metrics/

August 2009

The second milestone focuses on both network and performance metrics.

• Possible problems of directory authorities assigning Stable and Guard flags (PDF, 2M from April 1, 2009).
• Simulation of the number of Fast, Stable, and Guard flags for changed requirements (PDF, 229K from April 11, 2009).
• Reducing the Circuit Window Size in Tor (PDF, 110K from August 19, 2009).
• Performance of Requests over the Tor Network (PDF, 2.8M from August 24, 2009).
• Analysis of Circuit Queues in Tor (PDF, 196K from August 25, 2009).