Elastic Management of Byzantine Faults Luciana Arantes

Tolerating byzantine faults on a large scale is a challenge: in particular, Desktop Grid environments sustain large numbers of faults that range from crashes to byzantine faults. Solutions in the literature that address byzantine failures are costly and none of them scales to really large numbers of nodes. This paper proposes to distribute task scheduling on trusted nodes in a Cloud network and to have these nodes assess the reliability of worker nodes by means of a reputation system. The resulting architecture is built for scalability and adapts costs to the workload associated with client requests.