Abstract. We introduce a notion of stochastic noninterference aimed at extending the classical approach to information flow analysis with finegrain information describing the temporal behavior of systems. In particular, we refer to a process algebraic setting that joins durational activities expressing time passing through exponentially distributed random variables, zero duration activities allowing for prioritized/probabilistic choices, and untimed activities with unspecified duration. In this setting unifying time, priority, probability, and nondeterminism, we highlight the expressive power of stochastic noninterference with respect to the existing definitions of nondeterministic and probabilistic noninterference. From this comparison, we obtain that stochastic noninterference turns out to be very strict and limiting in real-world applications and, therefore, requires the use of relaxation techniques. Among them we advocate performance evaluation as a means for achieving a reasonable balance between security requirements and quality.
Fine-grain Models and NoninterferenceInformation flow analysis is a basic approach to the verification of security properties of systems which, in general, require to control who has access to what and when. Among the several conditions that describe the characteristics of unauthorized information flows, called covert channels, one of the most interesting, for its intuitive and wide-used idea, is the noninterference requirement [11]. Very briefly, in a multi-level secure system, the user at the high security level should not be able to affect what the user at the low security level can observe. Independently of the specific formalization of this notion, the underlying approach is based on the idea that checking noninterference is actually checking the indistinguishability of the different low-level views of the system that are obtained by changing the high-level behavior, see e.g. [22,9,17] and the references therein.Along the 90s generalized notions of noninterference were designed to analyze deterministic and nondeterministic systems. However, it was immediately clear that moving to a quantitative framework including fine-grain information, such as probability distributions associated with event execution, augments the distinguishing power of the observer [14,18,19,3]. More recently, the awareness that perfect noninterference is difficult to achieve has urged to estimate the information flow by employing this quantitative information and a relaxed notion of