[Bitcoin-development] P2P non-blockchain message proposal
john.dillon892 at googlemail.com
Mon May 13 08:19:03 UTC 2013
-----BEGIN PGP SIGNED MESSAGE-----
Peter's "Coinbase TxOut Hashcash" scheme mentions in passing anti-DoS
protection on the P2P flood-fill network for non-transaction messages, and an
application to use those messages, trust-free mixing. I did some review of the
source code and I think we can create a generalized P2P flood-fill message
system without difficulty.
First of all like Mike Hearn suggested in the tx-replacement (satoshi's
version) thread we can easily prioritize, or I should say deprioritize
non-essentially traffic with a simple "allocated bandwidth" scheme where nodes
set how many KB/second of bandwidth they want to give to P2P network messages.
Next out of that allocation use a priority scheme where higher priority
messages get put ahead of lower priority ones in the queue for retransmission.
Messages of too low a priority are dropped, and in general they basically just
don't get good network propagation, much like transactions with dust outputs
will increasingly face.
How do you determine priority? Why simply a genuine bitcoin sacrifice!
Previously that would have been some big bulky fidelity bond scheme, but now we
can use Peter's Coinbase TxOut Hashcash, or PowPos (proof-of-work
proof-of-sacrifice) as he mentioned in the forums. (and privately) I'll jump
the gun a bit and call it PowPos, it's a nice name as you say. :) We already
relay transactions based on 1mBTC/KB, so a similar per KB message cost is
Peter suggested using PayWords to amortize the cost of transmitting the PowPos
initial proof, which of course needs to be paid for. Storing original proofs
should be pretty cheap, so lets make the default to store them on disk, like
the UTXO set, with per-node settings for just how many of these proofs we are
willing to store. We can treat them kinda like accounts, and old enough
accounts simply get deleted, with a per-node configuration on how old is too
old. It's best effort, not permanent like the UTXO set. Nodes should be able to
ask their peers for the actual proof corresponding to a payment attempt if they
need it. (maybe a general delta compression system could do this and other
Nitty gritty: define either NODE_P2PMSG for "supports P2P messages" or a more
general NODE_SELECTIVE_RELAY to advertise general limitations on what a node
relays. (min txout value, replace-by-fee, p2p message etc?)
Define MSG_FLOOD_MSG inventory type. The logic is that flood-fill messages
*are* something you can have in your "inventory" of data, even though they are
still things that expire. Again nodes can set message expiration and prune them
in some sane way, a days worth of messages is probably fine for the vast
majority of nodes.
Each flood fill message needs a header with the sacrifice proof, either a full
PowPos, or a PayWord linked to a PowPos, followed by a type disambiguator
(64-bit uuid?) followed by the data payload. Limiting the payload to 100K by
default seems fine to me, same limit as on transactions.
The PowPos initial proof could also be it's own message type, MSG_POWPOS, and
again it's something you would have in your "inventory"
I haven't written up a formal trust-free-mix message proposal. Peter do you
have one? Seems to me negotiating transaction fees could be a bit tricky as you
want people to somehow at least say that yes they are willing to pay
transaction fees of X for the outputs they want, without revealing what those
inputs are. At the same time opportunistic mixing is cool, where you say "make
these coins available to be mixed if it doesn't cost me anything" and duplicate
other's txout values on the fly to make determining what is what as difficult
as possible, as well as occasionally broadcasting requests for random, or even
*common* txout values. (don't forget about zipfs law!)
Anti-anti-propagation DoS prevent is going to be hard too. You can rank nodes
based on the sum of genuine Bitcoin, or in this case genuine sacrifice activity
they are broadcasting to you. Tricky though because PayWord sacrifices aren't
stored anywhere, so a peer could reuse the same PowPos if they control all your
peers that are using the protocol.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
-----END PGP SIGNATURE-----
More information about the bitcoin-dev