Detail

Reliability in BFT Systems – Safe but not Always Live

The overall correctness of large-scale systems composed of many
groups of replicas executing BFT protocols scales poorly with the number of
groups. This is because the probability of at least one group being
compromised (more than 1/3 faulty replicas) increases rapidly as the number
of groups increases. In this paper we address this problem with a simple
modification to Castro and Liskov’s BFT replication that allows for
arbitrary choice of n (number of replicas) and f (failure threshold). The
price to pay is a more restrictive liveness requirement, and we present the
design of a large-scale BFT replicated system that obviates this problem.