Probablistic Logic Modeling of Network Reliability for Hybrid Network Architectures

Sandia National Laboratories has found that the reliability and failure modes of current-generation network technologies can be effectively modeled using fault treebased probabilistic logic modeling {PLM) techniques. We have developed fault tree models that include various hierarchical networking technologies and classes of components interconnected in a wide variety of typical and atypical configurations. In this paper we discuss the types of results that can be obtainedfvom PLMs and why these results are of great practical value to network designers and analysts. After providing some mathematical background, we describe the “plug-and-play ” fault tree analysis methodology that we have developed for modeling connectivity and the provision of network services in several current-generation network architectures. Finally, we demonstrate the flexibility of the method by modeling the reliability of a hybrid example network that contains several interconnected ethernet, FDDl, and token ring segments. communications networks and the time-dependent interactions between components [ 11 have been difficult to model with the fault tree, event tree, and reliability block diagram models that have been used successfully in other industries. However, network designers and analysts could benefit greatly from the information that PLMs can yield. An interdisciplinary team at Sandia National Laboratories has found that, in general, current-generation network technologies can be modeled using PLM techniques. We have developed a “plug-and-play” fault tree analysis methodology for modeling connectivity and the provision of network services in a wide variety of currentgeneration network architectures (e.g., ethernet, token ring, and FDDI) containing various types of components (e.g., routers, concentrators, MAUs, CAUs, servers, and workstations) interconnected in a wide variety of typical and atypical configurations. This paper describes that modeling technique and illustrates why the results that can be obtained from PLMs are of great practical value to network designers and analysts. 2. Benefits of Probabilistic Logic Models 1. Background For many years, probabilistic logic modeling (PLM) techniques have been used to help assess the reliability of complex electromechanical systems ranging from individual components within automobiles to large precision machine tools and even complex semiconductor fabrication facilities. Related techniques have been used to assess the risks associated with potentially high-consequence facilities such as chemical processing plants and nuclear power reactors. These techniques provide designers and analysts with key insights that can be used to predict the most important failure modes and vulnerabilities in the system. They can also provide quantitative guidance as to the most costeffective ways to improve overall reliability. While PLMs have been commonly used in many industries, their use in the telecommunications industry has been fairly limited. The complex topologies of PLMs have been used by a number of different disciplines, including quantitative reliability analysis (QRA), probabilistic risk analysis (PRA), and probabilistic safety analysis (PSA). Regardless of the discipline, the reasons for developing a PLM are the same: to identify an exhaustive list of the modes by which a system can fail, to find an approximate frequency with which we might expect to observe failures, and to determine a rank ordering of the components in the system by their “importance” to the proper function of the system. The “importance” of a component can be defined in a number of ways, but is often thought of as answering one of the following questions: How sensitive is the overall system reliability to changes in the reliability of each individual component? If the reliability of this component is allowed to decrease (say, by substituting components of lesser quality), how much will this affect overall system reliability?