Search results

The purpose of this paper is to describe a novel diagnosis approach, using neural networks (NNs), which can be used to identify faulty nodes in distributed and multiprocessor systems.

Based on a literature‐based study focusing on research methodology and theoretical frameworks, the conduct of an ethnographic case study is described in detail. A discussion of the reporting and analysis of the data is also included.

This work shows that NNs can be used to implement a more efficient and adaptable approach for diagnosing faulty nodes in distributed systems. Simulations results indicate that the perceptron‐based diagnosis is a viable addition to present diagnosis problems.

This paper presents a solution for the asymmetric comparison model. For a more generalized approach that can be used for other comparison or invalidation models this approach requires a multilayer neural network.

The extensive simulations conducted clearly showed that the perceptron‐based diagnosis algorithm correctly identified all the millions of faulty situations tested. In addition, the perceptron‐based diagnosis requires an off‐line learning phase which does not have an impact on the diagnosis latency. This means that a fault set can be easily and rapidly identified. Simulations results showed that only few milliseconds are required to diagnose a system, hence, one can start talking about “real‐time” diagnosis.

The paper is first work that uses NNs to solve the system‐level diagnosis problem.

The purpose of this paper is to describe an adaptive approach for diagnosing faulty nodes in a wireless mobile environment.

Based on a diagnosis approach that has been previously developed for wired networks. A discussion of the novel diagnosis protocol and its correctness is also included.

In this paper, the author presents a new implementation of a failure detection service for wireless ad hoc and sensor systems that is based on an adaptation of a gossip‐style failure detection protocol and the heartbeat failure detector. The authors show that the failure detector is eventually perfect – that is, there is a time after which every faulty mobile is permanently suspected by every fault‐free host, and no host will be suspected before it crashes.

The main limitations of this work is the lack of simulations results. Future works will be directed at providing such performance evaluations.

The work introduced in this paper aims mainly at identifying faulty nodes in a wireless mobile environment. This service has been identified as one of the basic building blocks for dependable mobile ad hoc networks (MANETs).

The paper is first work that adapts a previously developed failure detection service to wireless and MANETs.