Survivability of networked systems.

摘要

Over the past several decades, computer and communication networks have changed the way we live with their extraordinary growth in size, scope, and complexity. As a consequence, the impact of failure and outage of a component, subsystem, or system in the networked infrastructure becomes ever-increasingly significant upon the occurrence of disasters and attacks. However, traditional dependability analysis does not provide a clear definition, quantifiable metrics, and evaluation methods of survivability, although the design and deployment of survivable networked systems is a common practice in today's critical applications.;Survivability is often used to denote different, but closely related, concepts and measures in the current literature. The relationship between survivability, reliability, availability, performability is anything but clear. In this research we propose a general survivability quantification approach, which has a unified definition, consistent measures, and is applicable to a wide range of architectures, applications, and failure/recovery scenarios. This analytic modeling framework provides a general evaluation process that leads to unambiguous and verifiable survivability assessment, which can be a formidable task using current system-specific definitions and methods.;Based on the proposed framework of survivability analysis, we have developed state-space based Markov chain and Petri net survivability models for networked systems such as cable modem broadband access network, distributed database with voting algorithms, local telephone access network, and wireless cellular network. These networked systems differ in the following aspects with regard to survivability modeling: (1) The measures of interest can be different. For example, availability and downtime cost are investigated in the study of cable modem termination system, while call blocking and dropping probabilities are analyzed in wireless cellular network. (2) The type of analytic models can be different. For example, Markov chain models are formulated to compare the survivability of various local telephone access network architectures, while Petri net models are developed to evaluate survivability of voting algorithms in distributed database with non-zero network delay and site processing time. (3) The specific modeling techniques can be different. The telecommunications switching system illustrated in the general framework has a composite survivability model constructed by combining the system's pure performance model and availability model. However, hierarchical modeling is adopted to keep the analytic models tractable in local telephone access network. The approximation is reasonable because the call interarrival time in the performance model is much smaller than the event interarrival time in the availability model. When a system is complex enough that even hierarchical modeling cannot avoid the largeness problem, fixed point iteration is used to capture a system's characteristics of interest by focusing on an autonomous entity of the whole system and modeling its dependencies on the rest of the system such as in the study of wireless cellular system survivability.