Defense of Cyber Infrastructures Against Cyber‐Physical Attacks Using Game‐Theoretic Models
Defense of Cyber Infrastructures Against Cyber‐Physical Attacks Using Game‐Theoretic Models
复制标题
DOI：
10.1111/risa.12362
复制
发表时间：
2016-04
期刊：
影响因子：
3.8
通讯作者：
N. Rao;S. Poole;Chris Y. T. Ma;Fei He;J. Zhuang;David K. Y. Yau
中科院分区：
文献类型：
--
作者：
N. Rao;S. Poole;Chris Y. T. Ma;Fei He;J. Zhuang;David K. Y. Yau
网络基础设施的运作依赖于网络组件和物理组件，这些组件可能会受到不同类型的意外和故意降级。在网络和计算基础设施的背景下，我们使用博弈论模型研究攻击者和防御者之间的战略交互，该模型同时考虑了网络和物理组件。攻击者和防御者优化各自的效用，表示为成本和系统条款的总和。首先，我们考虑一个布尔攻防模型，其中网络和物理子基础设施可以作为单独的单元进行攻击和加固。其次，我们考虑一个组件攻防模型，在该模型中，其组件可以被攻击和防御，并且基础设施需要最少数量的攻击和防御才能发挥作用。证明了均匀费用下的纳什均衡在多项式时间内是可计算的，并为基础设施的生存提供了高层次的确定性条件。当攻击和防御成功的概率以及偶然失败的概率被纳入模型时，结果对攻击者有利，但在其他方面保持定性相似。这一方法已被我们在UltraScience Net基础设施方面的经验所激励和验证，该基础设施旨在支持高性能网络实验。然而，分析结果更具一般性，我们将其应用于云和高性能计算基础设施的简化模型。
The operation of cyber infrastructures relies on both cyber and physical components, which are subject to incidental and intentional degradations of different kinds. Within the context of network and computing infrastructures, we study the strategic interactions between an attacker and a defender using game‐theoretic models that take into account both cyber and physical components. The attacker and defender optimize their individual utilities, expressed as sums of cost and system terms. First, we consider a Boolean attack‐defense model, wherein the cyber and physical subinfrastructures may be attacked and reinforced as individual units. Second, we consider a component attack‐defense model wherein their components may be attacked and defended, and the infrastructure requires minimum numbers of both to function. We show that the Nash equilibrium under uniform costs in both cases is computable in polynomial time, and it provides high‐level deterministic conditions for the infrastructure survival. When probabilities of successful attack and defense, and of incidental failures, are incorporated into the models, the results favor the attacker but otherwise remain qualitatively similar. This approach has been motivated and validated by our experiences with UltraScience Net infrastructure, which was built to support high‐performance network experiments. The analytical results, however, are more general, and we apply them to simplified models of cloud and high‐performance computing infrastructures.