Maximization of Network Robustness Considering the Effect of Escalation and Accumulated Experience of Intelligent Attackers

網路的日益普及，帶來了日常生活上的便利，卻也伴隨而來更多的網路犯罪，因此網路安全及其強韌性之衡量已逐漸受到重視；對網路營運者而言，如何能有效的評估攻擊者行為及威脅也已日趨重要。 在本篇論文中，我們提出一個兩階的數學規劃模型來描繪網路攻防情境以及攻擊者行為；其中內層問題，我們探討攻擊者欲利用最小攻擊成本來攻克網路上多個核心節點，而在其攻擊過程中，會不斷的累積攻擊經驗，使未來的攻擊成本有效的降低；此外，在攻擊者攻克某一節點後，亦可在此節點上進行權限提昇，如此攻擊者便可擁有足夠的權限來探測更多此節點上所擁有的資訊；在此，亦衡量這些資訊可能會對網路所造成的影響，亦即，攻擊者在攻克多個核心節點時，會同時讓這些資訊所造成的影響，達到一定程度的傷害；而在外層問題中，目標網路的管理者則能有效配置其有限防禦資源，使攻擊者需花費的攻擊成本最大化。為了求得此問題的最佳解，我們採用以模擬退火法為基礎的演算法來處理此問題，並設計出多種不同的初始解以及尋找鄰近解的方法，藉此獲得近似最佳解。Internet has become much more important and worldwide, but it gives cyber criminals opportunities to crash a network system and conduct other cyber-crimes. Therefore, the issues of network security and robustness have come into notice. It is necessary for a network operator to understand the attacker behavior in order to efficiently allocate his limited budget. In this thesis, we propose a two-level mathematical programming model to describe the network attack and defense scenario. In the inner problem, an attacker’s objective is to compromise multiple core nodes using the minimum total attack cost. During the attack actions, the attacker may gain some experience from previous attacks to further reduce the attack costs in the future. Moreover, he can also pay extra fee to escalate on a compromised node to get higher user privileges, so that he will have higher authority to access more information on the node. We also measure the impact incurred by such information leakage in our model. As a result, the attacker will try to compromise multiple core nodes and collect valuable information, so that the total impact incurred by information leakage will exceed a threshold. Meanwhile, in the outer problem, the network operator of the target network allocates limited defense resources appropriately to maximize the total attack cost of the attacker. We adopt some Simulated Annealing-based algorithms to solve the problem and develop some initial solutions and several kinds of methods for searching neighbor solutions.口試委員審定書 I誌 II文摘要 IIIHESIS ABSTRACT IVable of Contents VIist of Figures IXhapter 1 Introduction 1.1 Background 1.2 Motivation 4.3 Literature Survey 8.3.1 Risk Management 9.3.2 Network Survivability 12.3.3 Attacker Behavior and Privilege Escalation 14.4 Proposed Approach 16.5 Thesis Organization 17hapter 2 Problem Formulation of the DRA and AEE Models 18.1 Problem Description and Assumption of the DRA Model 19.2 Problem Formulation of the DRA Model 28.3 Problem Formulation of the AEE Model 33hapter 3 Solution Approach 37.1 Simulated Annealing Method 37.2 Solution Approach for the AEE Model 41.3 Solution Approach for the DRA Model 47hapter 4 Computation Experiments 50.1 Computation Experiments with the AEE Model 50.1.1 Simple Algorithms 50.1.2 Experiment Environment 52.1.3 Experiment Results 59.1.4 Discussion of Results 73.2 Computation Experiments with the DRA Model 78.2.1 Experiment Environment 78.2.2 Experiment Results 80.2.3 Discussion of Results 84hapter 5 Conclusion and Future Work 87.1 Conclusions 87.2 Future Work 89eferences 91歷 9