Types of consensus mechanisms

All consensus mechanisms are developed to deal with faults in a distributed system and to allow distributed systems to reach a final state of agreement. There are two general categories of consensus mechanisms. These categories deal with all types of faults (fail stop type or arbitrary). These common types of consensus mechanisms are as follows:

• Traditional Byzantine Fault Tolerance (BFT)-based: With no compute-intensive operations, such as partial hash inversion (as in Bitcoin PoW), this method relies on a simple scheme of nodes that are publisher-signed messages. Eventually, when a certain number of messages are received, then an agreement is reached.
• Leader election-based consensus mechanisms: This arrangement requires nodes to compete in a leader-election lottery, and the node that wins proposes a final value. For example, the PoW used in Bitcoin falls into this category.

Many practical implementations of consensus protocols have been proposed. Paxos is the most famous of these protocols. It was introduced by Leslie Lamport in 1989. With Paxos, nodes are assigned various roles such as Proposer, Acceptor, and Learner. Nodes or processes are named replicas, and consensus is achieved in the presence of faulty nodes by agreement among a majority of nodes.

An alternative to Paxos is RAFT, which works by assigning any of three states; that is, follower, candidate, or leader to the nodes. A leader is elected after a candidate node receives enough votes, and all changes then have to go through the leader. The leader commits the proposed changes once replication on the majority of the follower nodes is completed. More detail on the theory of consensus mechanisms from a distributed system point of view are beyond the scope of this chapter. Later in this chapter, however, a full section is dedicated to the introduction of consensus protocols. Specific algorithms will be discussed in chapters dedicated to Bitcoin and other blockchains later in this book.