Learning from experience: Better design techniques for an improved consensus protocol

Active replication is a canonical approach for improving fault-tolerance of distributed server systems. Building abstract processes with signal-on-fail semantics is a well-known solution used to design dependable systems. Fail-Signal is a relatively newer abstraction recently presented to design efficient consensus algorithms. This paper critically analyzes an earlier work that proposed a primitive fail-signal based consensus protocol. The aim is to highlight various problems that will arise when the simplifying assumptions used in the design are excluded. Such analysis leads to some basic design choices for a more sophisticated protocol. It also provides basis for construction of Fail-Signal (FS) process with weaker set of assumptions.