Why high throughput?


#1

Lots of PoS based BFT algorithms are claiming to have very high rates of throughput*, combined with low latency. Much more so than what one finds in deployed Nakamoto consensus chains.

What critical tradeoff or assumption is made in these algorithms which unlocks this capacity improvement?

*I mean high rates on the base layer, not about enabling sharding or any L2 solution.


#2

Correct me if i’m wrong:

With POW, when a block is mined (miners)
Compute the new block ASAP => hash the hell out of it => Be the first to propagate it
With POS, when a block is mined
Proposers compute the new block => Validators validate and vote it => it gets part of the canonical chain

With POW, you are incentiviced to make that time to hash the block as long as possible to have a chance to find the solution to the puzzle. So you need the computation of the new block to be as fast as possible.
With POS you can take all that time to compute transactions, since it doesn’t matter how much time you spend as long as you vote in time and your vote gets propagated.


#3

If they are not sharding or partitioning the protocol in any way, these chains are likely making sacrifices in decentralization (requiring network participants to have larger than O© resources where C is the resources of a consumer laptop) and not making in protocol guarantees about data availability.

There are some gains to be had wrt more strict timing of events (not relying upon the random distribution of blocks via pow) but these are generally minor.


#4

Hi, thanks for the reply.

I am afraid I really don’t understand what you are saying, even though your descriptive premise of POS/POW seems about right, in particular what does this mean?

you are incentiviced to make that time to hash the block as long as possible to have a chance to find the solution to the puzzle


#5

Could you possibly elaborate on what these are?

This was my intuition also, however, I am seeing claims about 100s of validators, across multiple continents, sub 10s finality and 100s of transactions per second. This just does not seem to be something I would expect normal Nakamoto type chains to be able to handle, but perhaps I am wrong.