Consensus layer future evolutions?

Question

I have seen a number of people discuss possible evolutions of the tezos LPOS consensus layer in the future to incorporate current research on BFT protocols (see here and here for intros). Among them there is Avalanche and Tendermint.

My naive understanding is that those research projects aim at building protocols which have the same or similar amount of security as PoW but without the requirement of expensive energy consumption.

What i would like to understand better is

• what security features would an implementation of these 2 protocols bring to Tezos that its current consensus layer does not offer ?
• are there benefits other than stronger security ? (I can think of faster finality as one possibility but not sure if there are other ?)
• what are the main technical roadblocks that can get in the way of implementing such schemes in tezos and proposing them one day as amendments ?

Answer 1

The terminology should be clarified a bit:

1. Nakamoto consensus (bitcoin), the Tezos consensus algorithm are also "BFT" algorithms, they achieve consensus in the presence of Byzantine faults.

2. So-called LPOS is not a "consensus" algorithm, it's a way of selecting who gets to participate in consensus.

3. I think what you're trying to capture with the category you refer to as "BFT" are algorithms with finality properties i.e. that achieve consensus on each block, and which are secure under partial synchronicity assumption.

In terms of what this can bring to Tezos, concretely:

1. Faster confirmation of transactions: instead of waiting for many blocks, a single block becomes sufficient to consider transactions confirmed.

2. Shorter block time: as the partial synchronicity assumption relieves the system from having to make worst case assumptions about the network.

3. Better behavior in case of a consensus breaking bug: if bakers ever disagree on the validity of a block the network stops instead of splitting. This gives everyone a chance to diagnose and fix the issue.

4. Simpler security analysis: these systems are easier to reason about, and the proofs are much more straightforward, this makes it easier to iterate on them.