Can you at this point already provide an estimate on the cost per byte of randomness?
I don’t have an exact estimate, but at a high level there are three routes that the cost could take. One option is if the arkworks host functions are enabled on the relay, and the other is if they’re not, and a third is a hybrid of the first two.
Assuming the host functions are enabled, then we only need to fetch a pulse from Drand and verification can be done trustlessly on-chain. There’s virtually no networking overhead since it can be done by a single worker. We only store the drand pulse in this case, so a signature (48 bytes), randomness (32 bytes), and the round number (8 bytes). I assume this would be very cheap, with the costs incurred being due to the on-chain verification of a single pulse along with the fees associated with XCM dispatching and such.
The other route would be if we don’t rely on the arkworks host functions and would require a bit more effort to build, but would result in a more trustless and decentralized solution. In this case, we would rely on an committee (e.g. the set of collators) to fetch pulses, verify them offchain, and then submit signatures that the pulse is valid. This reduces trust in the system even beyond verification of drand pulses (by that I mean we don’t rely on an OCW to fetch pulses, since it isn’t guaranteed to execute anyway), however it carries greater computational and storage costs. This second approach is most likely the “right” approach. I don’t have an exact estimate, but it would be a higher cost than (1), but with less trust.
I think ideally a combination of both solutions would be my preferred approach, where workers fetch pulses from Drand, produce sigs on them that they gossip between each other, then once 2/3rds produce sigs on a pulse some chosen leader submits it on-chain to be trustlessly verified (so collators don’t do this offchain). The “IDN Beacon” discussed in the post uses DLEQ proofs for signature verification, so we can also expect for the solution to become ‘cheaper’ in the future.