Design of Cyber Security for Critical Infrastructures: A Case for a Schizoid Design Approach

Abstract

In this invited talk, we argue that designing cyber security of critical infrastructure requires a spilt-personality approach to design as opposed to design for correctness or for performance. Designing a functionally correct system, or a performance constrained system is fundamentally different in the sense that such design requires us to build models and to systematically refine models towards implementation such that correctness is preserved between refinements, and performance optimizations are introduced during refinements. Designing systems with cyber-security properties requires us to not only build models from theoretical principles, but also require modeling possible behaviors of an adversary. Modeling adversarial behavior is akin to test-driven model refinement, and hence not so different from certain approaches used when our goal is functionally correct design. However, for cyber-physical systems, we often need to detect an ongoing cyber attack since safe guards for cyber security often depend on assumptions which can be invalidated (e.g., insider attacks may invalidate perimeter security assumptions). Detecting ongoing attacks requires detecting behavioral anomalies in the physical system under cyber control – thus requiring us to build models from data. Machine learning approaches could be used to build such models. This we view as a schizoid approach – since the designer has to not only model the system from physical principles, he/she also has to build nominal behavioral models from data. While arguing this point of view, we introduce a virtual SCADA (supervisory control and data acquisition) laboratory we have built to help design cyber security of critical systems. The majority of this talk focuses on describing this software based virtual laboratory called VSCADA. Most of this research is published in [8,11] and summarized here for the sake of exposition to the present audience.