Emin Gün Sirer [ARCHIVE] on Nostr: 📅 Original date posted:2014-07-19 📝 Original message:> Most things I've seen ...
📅 Original date posted:2014-07-19
📝 Original message:> Most things I've seen working in this space are attempting to minimize
> the data transfered. At least for the miner-interested case the round
> complexity is much more important because a single RTT is enough to
> basically send the whole block on a lot of very relevant paths.
Agreed. Yaron's scheme is magical because it is non-interactive. I send you
a packet of O(expected-delta) and you immediately figure out the delta
without further back and forth communication, each requiring an RTT.
> I know much better is possible (see up-thread where I linked to an old
> proposal to use forward error correction to transfer with low data
> transfer (but not optimal) and negligible probability of needing a
> round-trip, with a tradeoff for more overhead for lower roundtrip
> probability).
FEC schemes are both fairly complex, because the set is constantly
changing, and (if i understand your suggestion correctly) they add
additional metadata overhead (albeit mostly during tx propagation). Set
reconciliation is near optimal.
In any case, I have no horse here (I think changing the client so it's
multithreaded is the best way to go), but Yaron's work is pretty cool and
may be applicable.
- egs
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linuxfoundation.org/pipermail/bitcoin-dev/attachments/20140718/d7cb8d29/attachment.html>;
📝 Original message:> Most things I've seen working in this space are attempting to minimize
> the data transfered. At least for the miner-interested case the round
> complexity is much more important because a single RTT is enough to
> basically send the whole block on a lot of very relevant paths.
Agreed. Yaron's scheme is magical because it is non-interactive. I send you
a packet of O(expected-delta) and you immediately figure out the delta
without further back and forth communication, each requiring an RTT.
> I know much better is possible (see up-thread where I linked to an old
> proposal to use forward error correction to transfer with low data
> transfer (but not optimal) and negligible probability of needing a
> round-trip, with a tradeoff for more overhead for lower roundtrip
> probability).
FEC schemes are both fairly complex, because the set is constantly
changing, and (if i understand your suggestion correctly) they add
additional metadata overhead (albeit mostly during tx propagation). Set
reconciliation is near optimal.
In any case, I have no horse here (I think changing the client so it's
multithreaded is the best way to go), but Yaron's work is pretty cool and
may be applicable.
- egs
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linuxfoundation.org/pipermail/bitcoin-dev/attachments/20140718/d7cb8d29/attachment.html>;