Fault diagnosis in a fully distributed local computer network

Graduation date: 1984Local computer networks are being installed in\ud diverse application areas. Many of the networks employ\ud a distributed control scheme, which has advantages\ud in performance and reliability over a centralized\ud one. However, distribution of control increases\ud the difficulty in locating faulty hardware elements.\ud Consequently, advantages may not be fully realized\ud unless measures are taken to account for the difficulties\ud of fault diagnosis; yet, not much work has\ud been done in this area. A hardcore is defined as a\ud node or a part of a node which is fault-free and\ud which can diagnose other elements in a system. Faults\ud are diagnosed in most existing distributed local computer\ud networks by assuming that every node, or a part\ud of every node, is a fixed hardcore: a fixed node or a\ud part of a fixed node is always a hardcore. Maintaining\ud such high reliability may not be possible\ud or cost-effective for some systems. A distributed\ud network contains dynamically redundant elements,\ud and it is reasonable to assume that fewer nodes are\ud simultaneously faulty than are fault-free at any\ud point in the life cycle of the network. A diagnostic\ud model is proposed herein which determines binary\ud evaluation results according to the status of the\ud testing and tested nodes, and which leads the network\ud to dynamically locate a fault-free node (a hardcore).\ud This diagnostic model is, in most cases, simpler\ud to implement and more cost-effective than the\ud fixed hardcore. The selected hardcore can diagnose\ud the other elements and can locate permanent faults.\ud In a hop-by-hop test, the destination node and\ud every intermediate node in a path test the transmitted\ud data. This dissertation presents another method\ud to locate an element with frequent transient faults;\ud it checks data only at the destination, thereby, eliminating\ud the need for a hop-by-hop test