Paper on aggregated vector consensus and Juno, an async common subset protocol with O(n^2) messages and a benchmarked implementation.
2 comments
Does the aggregation step mainly cut down the signature work, or is it also what makes the async common subset part cheaper to run with so many messages?
mostly the latter, i think, though the aggregation can also hide some per-node verification overhead depending on the concrete threshold scheme. the bigger win for acs is that it collapses the all-to-all value exchange into one aggregated vector per round, so the async part stops paying a separate coordination cost for every proposal.