Computer systems are becoming pervasive as well as growing rapidly in scale, complexity, distribution, and heterogeneity. For these reasons, it is becoming increasingly important for engineers to reason quantitatively about the various systems' properties of interest and curb their design, development, and deployment costs. In an ideal situation important system characteristics, such as performance and reliability, would be assessed at system design time, before significant time and cost have been devoted to a project. However, making useful (quantitative) predictions in early design stages is difficult at best, due ti the interplay between many relevant factors, such as complex properties of software components, the potential effects on software of the firmware (hardware, OS, device drivers), as well as the potentially conflicting desired system attributes.
Attacking even a small subset of these problems is challenging enough and would result in significant advances to the state of the art in complex systems engineering. Hence, this project proposes to focus efforts on design-time evaluation of architectures with respect to one key attribute - reliability. Here, reliability is defined as the probability that the system will perform its intended functionality under specified design limits. The proposed approach will enable an engineer to build a multi-faceted, hierarchical model of a system and assess its reliability in an incremental, scalable fashion. Although several software reliability techniques exist, they are insufficient. To address these deficiencies, the project will develop a technique that will couple software architectural models (well understood by system designers) with augmented Hidden Markov Models (which allow us to reason about numerous uncertainties existing in early design phases), and will augment this methodology with the relevant attributes of the firmware in support of more complete and meaningful reliability models.
The project will evaluate the results the methods developed along two measures of interest: tractability (intended to address scalability issues existing in real, complex systems) and sensitivity (intended to address issues of confidence in the researchers predictions under numerous uncertainties existing at design time) and will apply the results to real problems from the domain of mobile robotics, a problem domain that is representative of many complex, distributed, and embedded systems.
