Hello, I saw this paper floating around on Twitter, I was wondering if it’s claims are true and how it compares to algorand from the algorand team’s perspective
Consensus is only one part of the throughput equation. Other part is practical engineering related block production and network propagation.
I have just glanced at the paper. So please take everything below with a big grain of salt.
The paper looks very legit. The authors claim to improve on Algorand* (which is Algorand without sortition) and on the state of the art of partially-synchronous rotating/random leader consensus. This looks like a very nice result.
However, full Algorand also has sortition as well as the very strong property that each step of the protocol is done by a fresh newly-chosen committee. This is what allows Algorand to scale to millions of nodes without significant performance degradation.
A question (that may well be already answered as I really just glanced at the paper) is whether the new protocol proposed (“Simplex”) could be combined with sortition.
I think they claim it does but the wording is confusing. What do you think
“A major concern when designing modern consensus protocols is that of communication complexity. A low communication complexity is essential for a scalable protocol. In the multicast model, Algorand [CM19] showed an elegant way to subsample the committee of voters to achieve polylog(λ) multicasts (or O(n) messages in the point-to-point model). We remark that their techniques also apply to our protocol. (If we additionally trade off optimistic responsiveness, Algorand [CM19] can also achieve adaptive security.)”
Indeed, that seems to indicate that the sortition mechanism can be used with Simplex.
How much of optimistic responsive ness do you think they might trade off with regards to sortition. More than algorand or less. And is sortition what gives algorand adaptive security?