[Paper Review] Towards Predicting Efficient and Anonymous Tor Circuits

Armon Barton, Mohsen Imani+2 more|arXiv (Cornell University)|May 5, 2018

Internet Traffic Analysis and Secure E-voting28 references7 citations

TL;DR

This paper proposes PredicTor, a path selection technique for Tor that uses a Random Forest classifier trained on real-time performance measurements to predict and select faster, less congested circuits. Evaluated via simulations and live experiments, PredicTor improves Tor performance by 11–23% over Vanilla Tor and 7–13% over CAR, while maintaining anonymity through a new metric, CLASI, which reveals location-based anonymity leakage missed by traditional entropy metrics.

ABSTRACT

The Tor anonymity system provides online privacy for millions of users, but it is slower than typical web browsing. To improve Tor performance, we propose PredicTor, a path selection technique that uses a Random Forest classifier trained on recent measurements of Tor to predict the performance of a proposed path. If the path is predicted to be fast, then the client builds a circuit using those relays. We implemented PredicTor in the Tor source code and show through live Tor experiments and Shadow simulations that PredicTor improves Tor network performance by 11% to 23% compared to Vanilla Tor and by 7% to 13% compared to the previous state-of-the-art scheme. Our experiments show that PredicTor is the first path selection algorithm to dynamically avoid highly congested nodes during times of high congestion and avoid long-distance paths during times of low congestion. We evaluate the anonymity of PredicTor using standard entropy-based and time-to-first-compromise metrics, but these cannot capture the possibility of leakage due to the use of location in path selection. To better address this, we propose a new anonymity metric called CLASI: Client Autonomous System Inference. CLASI is the first anonymity metric in Tor that measures an adversary's ability to infer client Autonomous Systems (ASes) by fingerprinting circuits at the network, country, and relay level. We find that CLASI shows anonymity loss for location-aware path selection algorithms, where entropy-based metrics show little to no loss of anonymity. Additionally, CLASI indicates that PredicTor has similar sender AS leakage compared to the current Tor path selection algorithm due to PredicTor building circuits that are independent of client location.

Motivation & Objective

• To improve Tor's performance by reducing latency and congestion through intelligent path selection.
• To address the limitations of existing path selection algorithms that fail to dynamically adapt to network conditions.
• To develop a new anonymity metric, CLASI, that captures client AS inference risks from location-aware path selection.
• To evaluate whether performance-optimized path selection compromises user anonymity, especially through network-level fingerprinting.
• To enable deployment of a scalable, secure, and efficient path selection system in production Tor networks.

Proposed method

• PredicTor employs a Random Forest classifier trained on real-time measurements of Tor circuit performance, including latency and bandwidth.
• The classifier predicts the performance of candidate circuits before they are built, favoring paths with lower predicted latency and congestion.
• Training data is collected via periodic measurement streams from a centralized or distributed authority, sent to clients every hour.
• Measurement circuits are obfuscated by randomizing destination domains and payload sizes to prevent detection and manipulation.
• CLASI is introduced as a new anonymity metric that quantifies an adversary’s ability to infer a client’s Autonomous System (AS) through circuit fingerprinting at the network, country, and relay levels.
• CLASI models an all-knowing adversary with full visibility into Tor’s connection graph and auxiliary data such as relay bandwidth.

Experimental results

Research questions

• RQ1 Can a machine learning-based path selection technique like PredicTor significantly improve Tor’s performance without degrading anonymity?
• RQ2 How does PredicTor’s dynamic adaptation to congestion and distance compare to existing schemes like CAR and Vanilla Tor?
• RQ3 To what extent do location-aware path selection algorithms leak client AS information, and can this be captured by existing anonymity metrics?
• RQ4 Does CLASI reveal anonymity risks that traditional entropy-based metrics fail to detect?
• RQ5 What are the practical deployment challenges for integrating PredicTor into the live Tor network?

Key findings

• PredicTor improved Tor network performance by 23% in Shadow simulations compared to Vanilla Tor and by 13% compared to CAR.
• In live Tor experiments, PredicTor achieved 7–13% median performance improvement during high congestion, outperforming Vanilla Tor.
• CLASI revealed significant client AS leakage for location-aware path selection algorithms, where entropy-based metrics showed minimal anonymity loss.
• PredicTor exhibited similar AS leakage to Vanilla Tor because it selects circuits independently of client location, avoiding location-based fingerprinting.
• The Gini coefficient and uniformity analysis showed that relay selection under PredicTor remains balanced, with minimal entropy loss compared to Vanilla Tor.
• Time-to-first-compromise for PredicTor is expected to be similar to Vanilla Tor, as guard selection remains unchanged.