TING

Version: 1.0 Released: Friday, November 6, 2015 Contact: Frank Cangialosi ([email protected])

0 | Overview

This is an implementation of the Ting technique as outlined in our IMC 2015 paper Ting: Exploiting and Measuring Latencies Between All Tor Nodes (https://www.cs.umd.edu/~frank/pubs/ting-imc15.pdf).

Ting can measure the latency between any arbitrary pair of active nodes in the Tor network quickly and accurately, without requiring extensive measurement infrastructure or participation from the end hosts.

1 | Setup

Ting requires running 1 Tor client (used to create circuits) and 2 Tor servers (W and Z, as described in Section 3.3).

To get started with the default settings, simply run

./configure

This will:

• install the necessary dependencies (listed in section 5)
• build the included versions of Tor
• generate default Tor configuration files
• start running the Tor client and servers as daemons
• generate a default Ting configuration file

If you'd like more control over all of the settings, or run into any issues during setup, please check out Section 4, which describes Ting's configuration parameters, and Section 5, which describes all necessary dependencies.

2 | Important Notes

In the current release, Ting requires that the Tor servers W and Z are publicly listed in the Tor directories. For the next release, we plan to modify Ting to be able to speak to W and Z without publishing this information to the directories.

The default Tor configuration files we generate restrict the exit policy of W and Z to only allow exiting to this machine's public IP address to minimize malicious use of your relays (we didn't do this initially, which led a visit from the FBI :) However, note that this does not mean your Tor relay won't be used at all by other Tor clients: since it is available on the public list, it could be chosen as an entry or middle node.

Althrough running W and Z helps to minimize the variance Ting experiences, it is not an explicit restriction, the math still works out (as described in Section 3.3). For now, if you would prefer not to run your own Tor relays (or are unable to), you can do the following:

• remove the Tor installation and setup from the configuration file
• modify the IP and Fingerprint of W and Z in the tingrc file to match two Tor relays. Note that Z's exit policy must allow exiting to your public IP address.
• run Ting as described below

3 | Using Ting

Before running Ting, you must start the echo server so that Ting will have something to connect to. To do this simply run:

./echo_server

It will listen on the port specified by "DestinationPort" in the tingrc file.

Ting has two main modes:

A. Single -- use ting much like ping to quickly check the latency between two relays. Relays can be identified by ip or fingerprint. We recommend that you use fingerprints as ips are not necessarily unique.

./ting [fingerprint1] [fingerprint2] [optional args...]
./ting [ip1] [ip2] [optional args...]

B. Pairwise -- use Ting to measure the latency between all pairs listed in an input file. Again these may be specified by fingerprint or ip, but fingerprint is recommended.

./ting --input-file [input_file] [optional args...]

The number of samples to use and all other relevant settings are read from the tingrc file in the current directory, however these can optionally also be specified from the command line. To see how these are named, simply run:

./ting --help

4 | Advanced Configuration

The default tingrc file should be sufficient for most use cases, however if you decide to run Tor on different ports, for example, you can easily adjust for this in the tingrc file. Each possible option is listed below, along with a description, possible values, and the default value. See the default tingrc for an example. All options are key value pairs, space-separated, stored one per line.

ControllerPort controller port exposed by Tor client that allows stem to talk to the Tor client, default: 9051

SocksPort port exposed by the Tor client that allows us to default: 9050

SourceAddr address of this machine, stored in results file simply for record-keeping purposes, default: public address of this machine

DestinationAddr address of echo server, default: public address of this machine

DestinationPort port of echo server, default: 16667

NumSamples number of samples to take for each circuit, default: 200

NumRepeats number of times to measure each pair, default: 1

RelayList [internet | FILE] where to find the list of Tor relays, options: internet -- download from onionoo.torproject.org FILE -- path to local FILE containing relays, default: internet

RelayCacheTime if downloading the relay list from the internet, how long to cache the list locally before downloading a new version, in hours, default: 1

W [IP],[FP] IP and Fingerprint of W relay

Z [IP],[FP] IP and Fingerprint of Z relay

InputFile [FILE] if specified, measure all pairs in file. if file is "random", keep picking random pairs from the list of currently active Tor relays. default: none

Remote [PORT] if specified, bind to PORT and wait. upon recieving message of the form "relay1 relay2", measure the latency and respond with the result in form "Xms"

SocksTimeout timeout for SOCKS connection, in seconds, default: 60

MaxCircuitBuildAttempts after this many attempts to build a circuit, fail and move on, default: 5

5 | Dependencies

Packages:

• Python 2.7+
• Stem module 1.1.1+
• SocksiPy module (included in release)
• Tor (included in release)
• A public IP address (other Tor relays must be able to connect to you)

In our tests, we used the following versions of Tor. Since Ting builds on fundamental properties of Tor, it should be possible to implement it atop any version, however we have only explicitly tested these two. If the Tor control protocol changes in a newer version of Tor, it may be necessary to tweak the communication between Ting and Tor.

• Client: Tor v0.2.3.25-patched (included in release)
• Servers: Tor v0.2.4.22 (included in release)