Distributed consensus in semi-synchronous systems

Abstract

The Distributed consensus problem assumes that all processors in the system have some initial values; the goal is to make all non-faulty processors agree on one of these values. This paper investigates the time needed to reach consensus in a partially synchronous model with omission failures. In this model, the processors have no direct knowledge about time, but the time between consecutive steps of each processor is always between two known constants c/sub 1/ and c/sub 2/; the ratio C=/sup c2///sub c1/ measures the timing uncertainty in the system. Moreover, messages are delivered within time d. This paper provides an improved protocol for the above problem. When the majority of the processors are fault-free, the protocol achieves consensus in time 3( phi +1)d+Cd, where phi is the actual number of faults in a specific execution of the protocol. This allows an increase in efficiency up to 25% over the existing protocol which requires time 4( phi +1)d+Cd.<>