Progressive Model Generation for Adaptive Resilient System Software

Software continues to be a weak link in our critical systems. A prudent operator should employ a defense-in-depth strategy whereby"safe"systems are still monitored to detect breaches and respond to them. Unfortunately, such monitoring is challenging in practice, since there is no universal pattern that characterizes misbehaving software. We will capture an application"s intended behavior as it is coded in an IDE. The behavior will be stored as a model, and will be captured using a combination of automatic program analysis and manual tuning. We leverage modeling languages developed at the University of Pennsylvania for the purpose of runtime verification. These languages provide two levels of information: an abstract description of a system"s high-level behavior, and a mapping from high-level behavior to the source-level variables and procedures that actually implement the system. These models will supply a runtime monitor with the information needed to both determine when behavior is abnormal and determine what low-level events need to be tracked to observe the critical behavior.