Efficient and robust adaptive consensus services based on oracles

Lívia SampaioFrancisco Vilar BrasileiroRaul Ceretta NunesIngrid Jansch-Pôrto

Due to their fundamental role in the design of faulttolerantdistributed systems, consensus protocols havebeen widely studied. Most of the research in this area hasfocused on providing ways for circumventing the impossibilityof reaching consensus on a purely asynchronoussystem subject to failures. Of particular interest are theindulgent consensus protocols based upon weak failuredetection oracles. Following the first works that were moreconcerned with the correctness of such protocols, performanceissues related to them are now a topic that hasgained considerable attention. In particular, a few studieshave been conducted to analyze the impact that the qualityof service of the underlying failure detection oracle hason the performance of consensus protocols. To achievebetter performance, adaptive failure detectors have beenproposed. Also, slowness oracles have been proposed toallow consensus protocols to adapt themselves to thechanging conditions of the environment, enhancing theirperformance when there are substantial changes on theload to which the system is exposed. In this paper we furtherinvestigate the use of these oracles to design efficientconsensus services. In particular, we provide efficient androbust implementations of slowness oracles based on techniquesthat have been previously used to implement adaptivefailure detection oracles. Our experiments on a wideareadistributed system show that by using a slownessoracle that is well matched with a failure detection oracle,one can achieve performance as much as 53.5% betterthan the alternative that does not use a slowness oracle.

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: