The goal is to analyze the design of systems and network architecture in the context of failures and performance. The rationale is that analysis tools, used early in the design process and systematically throughout the different stages of that process, would be key enablers in reducing design costs, cost of change due to varying requirements, and verification and validation (V&V) costs, and would also increase reusability of network design with appropriate design changes for different ground vehicles or aircrafts.
Given this objective, we provide an approach to analyze network system architectures in terms of performance (latency/jitter/timing properties, bandwidth, buffer and other resources) and failure (fault modes, propagation) in a conjoint manner. In the past, such system analyses were done along one of those two dimensions in isolation due to the complexity of analyzing them together. The rationale behind this task is the insight that the trade-offs in the network architecture design space can be comprehensively explored only when dimensions are systematically explored in conjunction. Further, both fault and performance requirements are irrevocably linked such that any design change in one dimension impacts the other. The linkage between these two dimensions is not always completely understood and not formally characterized or analyzed. This has led to point solutions in the network architecture design space whereby, though an individual network architecture is designed to satisfy the requirements known at the time of initial design, any subsequent changes to the requirements in the later stage forced the design to be reworked ground up. Further, the need to integrate more applications on the same network or to leverage new hardware/software enhanced capabilities or adhere to limitations of the available technology since the time of initial design or even reusing the same network design for different vehicles/aircrafts with different requirements forced network designers to go to the drawing board and begin all over again with a new design. The analysis framework described here is designed to reduce network change cost and redesign cost by composing the network architectures to be systematically analyzed in terms of system requirements and thereby enabling network reuse as well as reuse of analysis artifacts (and thereby V&V artifacts).
We developed formal specifications of failure modes and their propagation based on the underlying protection mechanisms existing in the network hardware and software that prevent such failures, systematic evaluation and exploring trade-offs of system design with a set of analysis tools, dataflow and fault-tolerance modeling (availability, integrity), synchronization overheads characterization, hardware versus software trade-offs (implementing these services in hardware versus software and partitioning them), and path and system redundancy and system replication strategies.
Co-optimization of fault tolerance and performance with verification proofs of the general space of architectures is intractable. Our approach is to provide feasible trade-off analysis and verification of selected points in the design space. We support design choices through network architecture design space characterization at network components, host, and application-level redundancy management. Our approach is novel in that we integrate latency, utilization, buffer size and fault tolerance analysis and it is scalable to vehicle-sized network architectures.
Related Publications
Related Posters
Tool Download