Security Evaluation of the Cyber Networks Under Advanced Persistent Threats

Luxing Yang, Pengdeng Li, Xiaofan Yang, Yuan Yan TangIEEE Access 2017 · OpenReview

Abstract

Advanced Persistent Threats (APTs) pose a significant challenge to network security due to their stealthy and continuous nature. This paper introduces the Generic Secure-Compromised-Secure (GSCS) model, a dynamical systems framework designed to quantify network security over time. By defining "Limit Security" ($S_L$) as the long-term expected fraction of secure nodes, the authors provide a rigorous metric for evaluating defense strategies. The study proves the global stability of the system and explores optimal resource allocation between prevention and recovery.

Introduction

Advanced Persistent Threats (APTs) are not typical "smash and grab" cyberattacks; they are stealthy, well-resourced, and continuous. Because APTs persist over long periods, traditional point-in-time security snapshots fail to capture the true risk profile of a network. A major gap exists in cyber defense: the lack of a rigorous way to measure network security that accounts for the ongoing tug-of-war between an attacker’s infection efforts and a defender’s recovery processes. This research addresses this gap by creating a metric that reflects the long-term health of a network under constant siege.

The GSCS Model

The researchers modeled the attack-defense process as an individual-level dynamical system using a set of ordinary differential equations. The proposed "Generic Secure-Compromised-Secure" (GSCS) model categorizes each node in a network as either "Secure" ($S_i$) or "Compromised" ($C_i$).

State transition diagram of the Generic Secure-Compromised-Secure (GSCS) model.

As shown in the model diagram, nodes transition between states based on specific rates:

  • Infection Rate: Nodes become compromised via external attacks (direct hits, represented by $\alpha$) or internal infection (spreading from neighbors, represented by $\beta$ and the adjacency matrix $a_{ji}$). The prevention technical level $\delta$ and resources $y_i(t)$ mitigate this transition.
  • Recovery Rate: Nodes return to a secure state via recovery processes (patching, re-imaging, etc.), governed by the recovery technical level $\gamma$ and resources $z_i(t)$.

Using the LaSalle Invariance Principle and Dini derivatives, the authors proved that the network eventually settles into a unique, globally stable equilibrium state, regardless of the initial infection levels. From this equilibrium, they defined Limit Security ($S_L$) as $1$ minus the fraction of compromised nodes. For instance, an $S_L$ of 0.8 implies that 80% of the network is expected to be secure in the long run.

Key Insights and Results

The framework provides several practical takeaways for network defenders:

  • Optimal Resource Allocation: Security is maximized when defense resources are split equally between prevention and recovery. Over-investing in one area at the expense of the other leads to lower overall Limit Security. Simulations showed a clear peak in security when the ratio of prevention-to-recovery resources is approximately 1.0.
  • Connectivity as a Liability: Mathematically, adding new edges to a network—making it more well-connected—actually decreases Limit Security. Increased connectivity provides more pathways for internal infections to propagate.
  • Global Stability: The system's global stability ensures that even after a massive infection spike, the network will naturally gravitate back to its predictable Limit Security level, provided defense parameters remain constant.
  • Sensitivity Analysis: Limit Security increases with higher technical levels of prevention ($\delta$) and recovery ($\gamma$), while it decreases with higher levels of external attack ($\alpha$) and internal infection ($\beta$).
  • Diminishing Returns: While increasing total defense resources always improves security, the benefit follows a curve of diminishing returns, helping organizations optimize their security spending.

Conclusion and Future Work

The GSCS model provides a robust mathematical foundation for evaluating network security under APTs. However, the current model assumes static strategies for both attackers and defenders. Future work could incorporate Optimal Control Theory or Dynamic Game Theory to model adaptive adversaries who change tactics in response to defensive measures. Additionally, while the theoretical proofs are general, scaling the evaluation of $S_L$ for massive enterprise networks remains a computational challenge.

Summary updated

Abstract

Advanced persistent threats (APTs) pose a grave threat to cyberspace, because they deactivate all the conventional cyber defense mechanisms. This paper addresses the issue of evaluating the security of the cyber networks under APTs. For this purpose, a dynamic model capturing the APT-based cyber-attack-defense processes is proposed. Theoretical analysis shows that this model admits a globally stable equilibrium. On this basis, a new security metric known as the equilibrium security is suggested. The impact of several factors on the equilibrium security is revealed through theoretical analysis or computer simulation. These findings contribute to the development of feasible security solutions against APTs.
BibTeX
@inproceedings{yang2017security,
  author = {Luxing Yang and Pengdeng Li and Xiaofan Yang and Yuan Yan Tang},
  title = {{Security Evaluation of the Cyber Networks Under Advanced Persistent Threats}},
  booktitle = {IEEE Access},
  year = {2017},
  url = {https://openreview.net/forum?id=sHBca89PfBJ},
}

Loading related content...

Figure