Introduction
CloudSEC is a real-time detection framework designed for edge-cloud environments. It leverages Evidence Reasoning Networks (ERN) to reconstruct attack paths by correlating system vulnerabilities with real-time security alerts. This approach allows for the identification of complex, multi-stage lateral movement even in scenarios with inconsistent network data or resource-constrained nodes.
Challenges in Edge-Cloud Security
Edge computing decentralizes processing, which introduces significant security challenges. Traditional models struggle with data ephemerality, as edge nodes often have limited memory and transient data, complicating historical forensic analysis. Furthermore, dynamic architectures with shifting network topologies make it difficult to establish reliable behavioral baselines. CloudSEC addresses these issues by focusing on "how" an attacker arrived and "where" they might go next in real-time.
The CloudSEC Framework
The architecture of CloudSEC involves several distributed components working in tandem across cloud and edge layers to monitor and analyze system activity.
Key components include:
- EventTracker: A lightweight agent residing in Virtual Machines (VMs) or containers that monitors system calls and user activities.
- AlertCorrelator: Located at the edge, this component ingests alerts from multiple Network Intrusion Detection Sensors (NIDS) and maps them onto the ERN.
- Evidence Reasoning Network (ERN): Models logical relationships between known vulnerabilities to define potential attack paths.
Logic and Reasoning
The ERN defines attack paths through logical structures. For instance, an attacker might require multiple vulnerabilities (AND) or have alternative paths (OR) to reach a target.
CloudSEC also employs a Timing Independent Evidence Chain Reasoning algorithm to handle out-of-order logs common in distributed systems. Additionally, the system can generate "Virtual Records" to bridge gaps in missing evidence, maintaining the attack chain's integrity while adjusting the overall confidence score of the detection.
Evaluation and Results
The framework was evaluated using three primary sources: LLDOS 1.0, LLDOS 2.0.2 (MIT Lincoln Laboratories), and the Treasure Hunt dataset (UC Santa Barbara).
Key results include:
- Attack Reconstruction: In the LLDOS 1.0 test, CloudSEC successfully identified all five phases of the attack with 100% confidence.
- Resilience to Missing Data: In the LLDOS 2.0.2 test, where some evidence was intentionally omitted, CloudSEC used its virtual record logic to maintain the chain with an 86% confidence score.
- Performance: The system handled approximately 78 to 120 security events per second, demonstrating its suitability for real-time deployment at the edge.
Conclusion
CloudSEC provides a robust, vulnerability-centric approach to lateral movement detection. While effective, future work is needed to address the dynamic generation of ERNs for hyper-dynamic environments and to detect "living off the land" attacks that utilize legitimate administrative tools rather than vulnerability exploits.
Summary updated