TL;DR
This paper introduces OD1SET, a novel on-line distributed system that uses autonomous agents to trace the origin of network connections, even when users employ complex "looping" techniques to obscure their path. By employing a loop-unwinding algorithm and secure communication protocols, OD1SET aims to provide real-time, efficient, and robust session tracing for security incident response.
What Problem Does This Paper Address?
In today's interconnected world, unauthorized access to computer systems is a growing concern. When security incidents occur, it's crucial to identify the source of the intrusion for legal, accountability, and future prevention purposes. However, tracing the origin of a network connection can be incredibly challenging. Malicious users can employ techniques like "IP hopping" or creating connection loops across multiple hosts to deliberately obscure their true starting point. This makes it difficult for system administrators, who often lack direct access to all involved machines, to pinpoint the original attacker. Traditional off-line methods relying on audit logs are vulnerable to tampering or deletion, especially in compromised systems. This paper tackles the problem of real-time, on-line session tracing in a distributed network environment, aiming to overcome the limitations of existing approaches.
Methods
OD1SET employs a multi-agent system where each agent resides on a host and collaborates to trace connections. The core of their approach is a loop-unwinding technique. When an agent detects a connection, it initiates a tracing process by communicating with other agents. If it encounters a repeated connection pattern (a loop), it identifies and "unwinds" it, effectively reducing the complexity of the tracing path and saving time. This is achieved by agents storing information about user sessions and connections in their knowledge bases.
To ensure secure communication between agents, OD1SET utilizes RSA public-key cryptography for initial negotiation and establishes symmetric session keys (DES) for efficient, encrypted data exchange. This prevents eavesdropping on agent communications.
Furthermore, to mitigate Denial of Service (DoS) attacks, the system implements a dynamic port-allocation scheme. Agents initially communicate on a well-known port to agree on a temporary, random port for their actual data exchange. This port is frequently switched, making it difficult for attackers to target a specific listening port.
Datasets
Not discussed in the paper. The experiments were conducted on a test facility consisting of seven Linux machines.
Key Insights and Ideas
- On-line tracing is feasible and necessary: The paper demonstrates that real-time tracing is achievable without relying on potentially compromised audit logs, using live system data instead.
- Loop unwinding is crucial for efficiency: The proposed loop-unwinding algorithm is key to handling complex, deliberately obfuscated connection paths, significantly reducing tracing time.
- Agent collaboration is powerful: Distributing the tracing task across multiple agents on different hosts allows for parallel processing and a more comprehensive view of the network.
- Security through obscurity (and encryption): Dynamic port allocation and robust encryption are vital for protecting the tracing agents themselves from attacks and ensuring the integrity of the tracing process.
- Lightweight agents are practical: The focus on a lightweight agent design is important for widespread deployment without imposing a significant performance burden on the network.
Key Results
The experimental results show that OD1SET effectively unwinds connection loops. The tracing time grows sub-linearly with the number of hops, indicating good scalability. While encryption adds overhead, the paper notes that for realistic connection chain lengths (e.g., up to thirty hosts), the performance cost is acceptable and necessary for security. For instance, with seven hosts and 32 connection hops, tracing time increased from 0.73 seconds without encryption to 3.99 seconds with encryption, a trade-off deemed worthwhile.
Limitations and Future Work
The authors acknowledge that OD1SET currently works exclusively with alive connections. If a user disconnects from a host, the agents lose visibility. They suggest that extending the system for off-line tracing is a possible future direction, though other systems already address this.
A significant limitation is that the system primarily detects users connected via interactive sessions that update system tables (like rlogin or ssh). If an intruder exploits a vulnerability to gain shell access without using these commands, they might go unnoticed.
Future work includes:
- Integrating OD1SET into an agent-based intrusion detection system.
- Developing more robust agent survivability mechanisms (replication, mobility, zero-loss).
- Extending the tracing capabilities to include dial-up connections.
Summary updated