RAFT Leader Election and Casper


#1

What is the disadvantage to using RAFT leader election for the validator committee in Casper whereby validators are moved in and out of the committee one at a time versus swapping out the entire committee at regular intervals?

This would allow us to leverage a tried and tested approach to leader election and would allow for a constantly fluctuating pool of the validator committee members vs. locking up the collective in for a predetermined number of rounds. We could perform the same sort of operation on the current committee for deciding which member(s) will be being replaced by the new committee member(s).


#2

RAFT is a protocol that works under assumption that all servers in the cluster are not malicious. This is not the case for decentralized networks, where it is assumed that a minority of cluster members are malicious.


#3

That’s a good point @kladkogex, assumption of node good withstanding in the network is not safe to assume. You may check out the concept of Node Trust Quotient used in the link: https://kipfoundation.github.io/techprimer/4-Sentinel-Scalability.html.

The proposed system runs on TARA - A modified version of Raft; and collects a set of information from hardware as well as dynamic statistics when required.

The math is not fully carved out yet, but I hope this adds value to the thread.


#4

Once you start carving out the math you will encounter the Byzantine generals problem …