A SCALABLE TI-LEVEL DISTRIBUTED SYSTEM-LEVEL DIAGNOSIS IN DYNAMIC FAULT ENVIRONMENT

Abstract

The dissertation considers the problem of achieving a scalable distributed diagnosis and performance tuning in dynamic fault environments. It is assumed that network nodes are subject to crash faults and the communication links are fault free The diagnosis approach assumes a distributed network, in which nodes can test other nodes and can determine them to be faulty or fault-free. In dynamic fault environment, no restriction is placed on the number of nodes that can fail or on the time of failure or recovery. However, the frequency with which events can occur on a single node is considered limited to achieve correct diagnosis. Each node in system does system diagnosis independently. Existing work is aimed to reduce either latency or network utilization but scale poorly. A new diagnosis algorithm, called Multi-level DSD, is shown to provide scalability, which controls both latency and network utilization in fully connected networks. The algorithm is scalable in the sense that only minor modification is required to diagnose system with large number of nodes. Furthermore, it is possible to tune performance of diagnosis according to system requirement. Multi-level DSD divides system in clusters of nodes, where each cluster is either a single node or group of clusters. Cluster diagnoses itself by running a diagnosis algorithm between its sub clusters called cluster diagnosis algorithm. Clusters at each level runs same cluster diagnosis algorithm. The diagnosis can be configured to adapt requirements through alteration of number of nodes and cluster diagnosis algorithms at different levels. The extension of algorithm is presented which diagnoses any arbitrary network. The algorithm has been simulated using OMNeT++ network simulator for varying size of networks.