Chained Tendermint: A Parallel BFT Consensus Mechanism

A number of well-known BFT (Byzantine Fault Tolerant) algorithms, such as Tendermint, Casper, HotStuff, and Grandpa, have been developed in recent years to solve the problem of consensus. Of them, Tendermint can ensure instant finality but needs multiple rounds to commit a block. Casper and HotStuff both need a constraint on the direct parent to commit to a block. Grandpa uses ghost function to extract a common block from a vote set, but it is not based on a chained model and does not use signature aggregation to boost performance. We propose a consensus algorithm based on the theories of Tendermint and HotStuff. It is based on the same network model (each message sent out can reach every replica eventually, the adversary can hold it but not forever) and adversary assumption (the adversary can do Byzantine behavior like not proposing block or sending vote, or sending contradictory votes) as in Tendermint, but does not have the feature of instant finality. The result is an increase in throughput, since the blocks are handled in the chained model, the processes of handling each block can overlap.