Quantification of Security and Survivability ITI Workshop on Dependability and Security Urbana, Illinois Kishor Trivedi Department of Electrical and Computer Engineering Duke University Durham, NC 27708-0291 Email: kst@ee.duke.edu Home page: www.ee.duke.edu/~kst
Outline Quantification of security Quantification of survivability
Security Quantification Security Attributes Integrity, confidentiality, availability Authentication, non-repudiation Threats Design, physical, interaction faults Attacks Security Evaluation Qualitative assessment Certain checklists as security evaluation criteria, tiger team Quantitative assessment Based on probabilistic approach
Related work Littlewood et al. explored the feasibility of probabilistic quantification on security Ortalo et al. used privilege graph to model system operational security Jha et al. used attack graph to model attacker behavior Singh et al. designed SANs (stochastic activity networks) model for probabilistic validation of security and performance of several intrusiontolerant architectures Chen et al. analyzed vulnerabilities using finite state machine model Jonsson et al. conducted experiments and presented a quantitative model of security intrusion based on attacker behavior Stevens et al. proposed probabilistic methods to model the DPASA (Designing Protection and Adaptation into a Survivable Architecture) architecture
Probabilistic Security Quantification Our research publications Security analysis of SITAR Intrusion Tolerant System, ACM Workshop on Survivable and Self-Regenerative systems, Oct. 2003 A method for Modeling and Quantifying the Security Attributes of Intrusion Tolerant System Performance Evaluation Journal, 2004 Security modeling and quantification of intrusion tolerant systems using attack-response graph, Trusted Internet Workshop, Dec. 2003 Our approach: design state transition diagram of system security states, and use Markov chains, Semi Markov Process, SRN and Attack Response Graph to develop high fidelity models incorporating both attacker and system behavior.
SITAR Overview SITAR is an intrusion tolerant architecture developed jointly by MCNC and Duke SITAR uses spatial redundancy, diversity and adaptive reconfiguration to achieve intrusion tolerance SITAR architecture Proxy modules (PM) Acceptance monitors (AM) Ballot monitors (BM) Audit control module (ACM) Adaptive reconfiguration module (ARM) COTS servers
Security Quantification of SITAR Security vs. attack rate 1 0.995 0.99 0.985 0.98 0.975 0.97 0.965 0.96 0.955 threat level 3 threat level 1 0.33 2.33 4.33 6.33 8.33 10.33 12.33 Threat level 1 System security 30000 25000 20000 15000 10000 5000 0 Mean time to security failure vs. attack rate 1 2 3 4 5 6 7 8 9 10 11 12 13 threat level 3 threat level 1 Threat level 3 Mean time to severe security failure
Security Quantification Challenges Appropriate modeling of cyber attackers Need to determine appropriate level of detail/abstraction Need different attacker models for different purposes and attack classes Comprehensive modeling of system-level security quantification Difficult to model certain security attributes such as confidentiality and integrity using probabilistic techniques Hard to comprehensively quantify high-level security requirement with different security attributes using a single approach Measurement techniques for model parameterization and validation Hard, careful work and significant time required for data collection
Survivability Quantification Threats Natural disasters Man-made accidents Hardware/software faults Malicious attacks Quantitative evaluation John Knight A survivability specification is a four-tuple {E, R, P, M}, E: operating environment; R: tolerable service; P: pmf on R; M: finite-state machine of state transition (analogous to availability). Soung Liew r-percentile survivability N r is the probability that N is no greater than r % of the total resource (analogous to performability). T1A1.2 working group Survivability depicts the time-varying system performance after a failure occurs
Our Survivability Research Analysis approach Develop, parameterize, and solve Markov and non-markov models including failure modes, traffic patterns, and resource contention. T1A1.2 based survivability measures do NOT depend on the disaster rate; this may be considered good as the disaster rate is hard to quantify in practice Our Publications Transient behavior of ATM networks under overloads IEEE INFOCOM 96, pages 978 985, San Francisco, CA, March 1996. Network survivability performance evaluation: a quantitative approach with applications in wireless ad-hoc networks ACM International Workshop on Modeling, Analysis and Simulation of Wireless and Mobile Systems (MSWiM' 02), Atlanta, GA, September 2002. A general framework of survivability quantification Proc. of l2th GI/ITG. Conf. On Measuring, Modelling and Evaluation of Computer and Communication Systems (MMB 04) Survivability analysis of telephone access network Proc. of 15th IEEE International Symposium on Software Engineering (ISSRE 04)
Availability and performance models λ λ λ 0 1 2 n-1 n µ 2µ 3µ nµ λ Pure performance model To compute blocking prob. In each state of the availability model Pure availability model
Force a failure in the system P bk =1 Survivability Model and Results Survivability results blocking probability P bk =0.0079366 T R : relaxation time P bk =1 P bk =1 P bk =1 P bk =0.0079366 T R Normal operation in this state Make this Absorbing state Make this the initial state
Excess Loss Due to Failure (ELF) ELF: a survivability measure reflecting the total loss before the system is recovered P bk (t) P bk (t=0) Area in the shadow Dropped calls + Excess blocked calls = ELF
Comparison: six proposed architectures of Telephone access network Active/standby IIC1 Active/standby IIC2 Active/active IIB I, IIA Relaxation time* 35310 s 35370 s 32940 s 118 days Call loss due to failure 9920 9920 4944 9920 Extra call loss due to blocking 19729 25323 15804 9*10 7 ELF 29649 35243 20748 9*10 7 III 0 s 0 0 0 Y. Liu, V. Mendiratta, K. S. Trivedi, Survivability analysis of telephone access network Proc. of 15th IEEE International Symposium on Software Engineering (ISSRE 04)
Survivability Quantification Challenges No unified definition Variation due to different metrics Steady state or transient Availability, capacity-oriented availability, or performance Variation due to different systems Wire-line/wireless access networks, optical transport networks, military 3C networks, financial and banking networks, etc. Increased system complexity Heterogeneity Components have different capacity, performance, fault tolerance Multiple layer hierarchy Cross layer dependence, fault propagation, resource allocation & optimization Failure scenario and impact Identify potential failures and their impact on services