The reliability and availability of distributed services can be ensured using replication. We present an architecture and an algorithm for Byzantine fault-tolerant state machine replication. We explore the benefits of virtualization to reliably detect and tolerate faulty replicas, allowing the transformation of Byzantine faults into omission faults. Our approach reduces the total number of physical replicas from 3f +1 to 2f +1. It is based on the concept of twin virtual machines, which involves having two virtual machines in each physical host, each one acting as failure detector of the other.

No abstract