view change algorithm and paxos

Question

I was wondering what is the relationship between view-change algorithm and Paxos? In my lecture notes it states that "participants in each view agree on the primary, that later on governs the replication process". What is the view in this context and how Paxos relates to this?

Answer 1

The concept of "view" in paxos is applicable in scenarios where the number of nodes (proposers/acceptors/learners) is dynamic.

Imagine a scenario where nodes discover each other via UDP broadcast on startup. At any point in time, you can have a variable number of nodes running. Each time this "number of nodes" changes, that would indicate a new view.

Why this is important is because in such a dynamic environment, how do you ascertain what the "quorom" size is for accepting messages? By prepending a monotonically increasing ViewNumber to the ProposalNumber (also called a Viewstamp), this allows for graceful detection of changes in the "view" or number of nodes.

The whole view/replication concept is coined in the article:

Paxos Made Practical