Reliability Analysis Of Two Terminals Network Based On Discrete Probability Model

As the economy society depends on communication networks more and more, the network failures will bring about a great data loss or even disaster. With the network reliability analysis, the engineers can enhance the network reliability and reduce the loss when the failures take place. Although the network reliability analysis has been discussed widely, some works need to be furthered:Firstly, the reliability analysis of network with imperfect nodes. The traditional assumption of perfect nodes is inappropriate for new fields such as ad hoc networks and some networks in disaster environment. Furthermore, the number of reliability computations will largely expand when there are many imperfect nodes.Secondly, the reliability analysis of network with dependent componets failures. Because the componets failures are usually assumed to be s-independent (statisticly-independent), the traditianl methods overestimate the network reliability. Furthermore, this assumption is inappropriate for networks in extreme environment. When the network scale and number of dependent failures are large, the network reliability analysis will become difficult.Thirdly, the importance analysis of network components. The tranditional methods mainly focus on the reliability values, and little reliability information can be known from the point of components failures. There needs some effective methods for the importance analysis of network components.Fourthly, the runtime reliability analysis about networks. The tranditional methods mainly discuss the reliability metrics for design, which can not indicate the reliability when the networks are operating.Because of the large scale and dynamic complexity, it is difficult to describe the network failures with those classic failure distribution functions. To simplify the complexity and resolve the general problem, this thesis makes use of current prevalent method--discrete probability models with time independence. Although olny the two terminals networks are discussed, the methods can be applied into the researches of K-terminal network and all terminal network.This thesis focuses on the network reliability analysis and carries out the following researches:(1) The reliability analysis about networks with imperfect nodes. Two enhanced OBDD (Ordered Binary Decision Diagram) method are presented, with the names of MFP (Marked Factor Partition) and SNR (Street Network Reliability). The MFP evaluates the network reliability based on the marked factor partition, and the SNR evaluates the network reliability based on character indentification. These methods can evaluate the reliabilities of some large networks with imperfect nodes. Otherwise, with consideration of many imperfect nodes, the EF (Enhanced Factoring) is presented to evaluate the reliability of WSN (wireless sensor networks).1) MFP enhances the reliability analysis method based on OBDD. During the OBDD construction, the ER (Edge Replacement) operations are executed to represent the imperfect nodes with OBDD nodes. Two points improve the computation efficiency for the large scale networks: during the decomposition, the marked factor partitions are used to identify the ismophic sub-networks and the repeated computations are decreased from these networks; the states of nodes and edges are stored in OBDD and the redundant states are decreased.2) The computation course of SNR is similar with MFP, but is is especial for the street networks. The difference is the identification of ismophic sub-networks, SNR makes use of the source node position to characterize each sub-networks and indentify those sub-networks with same structures.3) EF enhances the factoring and improves efficiency with hash table, which stores the reliabilities of ismophic sub-networks.(2) The reliability analysis about networks with CCF (Common Cause Failure). Based on the model of common cause failure, the CNR (Common-cause-failure Network Reliability) and WR (Wireless-sensor-networks Reliability) are presented. The CNR is for a general networks with CCF, and it can analyze the large networks with many common cause events (CCE).1) The CNR is an enhanced OBDD algorithm, and the nodes are assumed to be perfect for simplification. During the reliability computations, only one recursive OBDD construction is executed. When the network scale and CCE number are large, this kind of OBDD construction greately decreases the running time. About the OBDD constructions for CCF networks, two methods are proposed: the first one is based on Boolean operation, and it is more flexible; the second one is based on common cause variable sets, and it is more storage efficient.2) With the considerations of large number of CCEs and imperfect nodes, WR can evaluate the reliability of WSN. The basic principle of WR is samiliar with CNR, but it effectively resolves the imperfect nodes problem with node expansion.(3) The importance analysis of network components. Some importance metrics are dicussed: Birnbaum measurement, Critical importance, Risk increment interval, etc. Two importance ananlysis methods for general networks are presented: BIL (Birnbaum Importance of Link) and RI (Risk Increment). The BIO computes the Birnbaum importance of links, and indicates the weakness and the fault component with the most probability. As an evaluation method for the risk of component failure, the RIO can also analyze the network stability, which can be used as invulnerability metric. Otherwise, based on the risk increment, the WNI (Wireless Node Importance) is presented to evaluate the node importance of WSN.(4) The reliability analysis based on network measurement. The reliability metrics are presented for fault-tolerant MPLS network and resilient IP network. The one for MPLS network is based on the number of failure, and the one for IP network is based on available bandwidth. For the runtime reliability analysis of MPLS network, an adaptive loopback mechanism for LSP failure detection is presented. It is a fast and low costs method for MPLS failure detection.