Um Detector de Defeitos Cumulativo Baseado em uma Abordagem Difusa

Joás E. SouzaFernando Castor

Failure detectors are fundamental components of large-scaledistributed systems. A failure detector must perceive node crashes promptlyand adapt a wide array of network conditions. In addition, it has to beaccurate and avoid wrong suspicions, since the latter result in unnecessaryprocessing and communication. Moreover, failure detectors should, ideally,have modest requirements in terms of memory consumption and processingtime. This paper presents an accrual and fuzzy failure detector (ACD). TheACD employs ideas from Fuzzy Logic and outputs a suspicion level on acontinuous scale, instead of simply considering a monitored process as"correct" or "suspect". Furthermore, the ACD requires less resources thanother accrual failure detectors. Our experiments show that, in some scenarios,the ACD performs better than well-known failure detectors described in theliterature and, in others, has a comparable performance.

Caso o link acima esteja inválido, faça uma busca pelo texto completo na Web: Buscar na Web

Biblioteca Digital Brasileira de Computação - Contato:
     Mantida por: