[Bitcoin-development] Near-block broadcasts for proof of tx propagation

[Peter Todd <pete@petertodd.org>]

[-- Attachment #1: Type: text/plain, Size: 7595 bytes --]

Currently there is no way for a node, SPV or otherwise, to know if a
transaction has been broadcast succesfully to any amount of hashing
power. This makes it difficult to determine if a transaction failed to
either propagate across the network, or failed to pay sufficient fees to
be worthy of inclusion in a block.

Broadcasting blocks that almost, but not quite, met the difficulty
target provides clients with fast and honest proof about the hashing
power mining their transaction. This proof is inherently honest because
making a "near-block" is an expensive operation; additionally given at
least one honest peer a node can detect near-block censorship by any
other peer statistically.


Limitations of fee estimation
=============================

Mempool-based fee estimation is limited by the ability of peers to lie,
particularly to SPV peers. Miners wishing to increase fees can conduct
sybil attacks where they lie to peers about the average fees required to
get transactions into blocks. This problem is particularly dangerous
given the lack of incentives to run full-nodes in the first place; the
number of full nodes has continued to drop over time as users switch to
SPV clients and web-wallets; it would be unfortunate if users started
switching to web-wallets because they could offer better fee estimation.
In any case creating monetary incentives to sybil the network is very
undesirable.

Out-of-band fee payment has the opposite effect of making fees in blocks
appear lower than actually required to get them mined; transactions will
get stuck unless an initial bad estimate can be replaced with a higher
paying one. The payment protocol makes out-of-band fee payment
attractive in the case where you want to accept a payment from a
customer and pay the fee for them; child-pays-for-parent wastes money
paying for additional blockchain space.

Replacement-based schemes allow for recovery from stuck too-low
transactions, but they are still succeptable to sybil attacks. (don't
relay the transaction to other pools)


Miner incentives to create near blocks
======================================

Why would a miner want to go to the trouble of broadcasting a near block
anyway? Wouldn't it be better if users didn't get feedback about fees
and over-paid instead?

If you are a large miner as a % of total profit efforts such as sybiling
the network have a greater rate of return; if you are a small miner the
greater income you receive from deception is outweighed by the cost.
Thus you have an incentive to provide mechanisms to force larger miners
to behave honestly.

Secondly near-blocks could help "pre-propagate" transactions that will
be mined in the near future, thus reducing block propagation times and
hence orphan rates. (the pre-propagation can use the proof-of-work to
rate-limit transactions that nodes would otherwise not forward, also
allowing non-full nodes to safely participate in relaying) Again, this
is something that most interests smaller miners with less peak bandwidth
rather than large pools.

In the event of a fork near blocks can be used to more quickly determine
which side of the fork has the majority of hashing power, allowing the
minority side to switch sooner. Again the reduction in orphan rates
benefits smaller miners more than larger ones. (though note how only
near-block headers are required for this application)


Contents of a near block
========================

From the miner incentives we see that near blocks should contain two
types of information:

1) Transactions known to the miner, but not included in the current
block. This information allows nodes to determine if a transaction they
have broadcast was succesfully propagated to the majority of hashing
power regardless of whether or not it is being mined, allowing nodes to
avoid sybil attacks attempting to censor the transactions they make.

This information needs to be committed separately in the coinbase tx.
The incentive for miners here is to ensure that no-one can gain an
unfair advantage with a sybil attack.

2) Transactions the miner is attempting to mine, proved by the merkle
root. The incentives here are allowing non-full nodes to safely
propagate transactions, improving block propagation, as well as further
preventing deception by larger miners.

Transaction mutability complicates both #1 and #2. In the case of #1
nodes can exploit mutability while relaying transactions, although at
least relaying mutability is increasingly difficult; the incentives are
such that the miners themselves have no reason to lie about the txids
they know of. For #2 the incentives are all harmed by mutating
transactions, so again we can expect miners to either leave transactions
as they are, or simply not publish near blocks at all.

Bandwidth usage is reasonable: the average transaction from the last
10,000 blocks is 450 bytes. Both data sets can be delta compressed
against previously sent txids. Even a naive implementation that sends
full txids would result in near blocks that are about 1/10th of the size
of full blocks. (32-byte txids, and 1/4 of that amount in the "seen but
not mined" list) The machinery for near blocks can also be easily
re-used to implement improved full-block relaying with transactions in
blocks being referenced to by txid.


Replacement with near blocks
============================

An node making a transaction can do the following:

1. Broadcast the transaction with an initial estimated fee. (the txid is
added to the bloom filter here) The estimate can safely be be on the low
side.

2. Wait

3. If transaction still hasn't appeared in either a block or near block,
rebroadcast with a higher fee, either by replace-by-fee method, or
zero-conf safe method of adding an additional txin+txout.

Peers practicing censorship of either transactions or near blocks can be
detected statisticly by preferring to connect to peers that provide more
near blocks. Note how a short 80-byte near block header is sufficient
information to detect a peer withholding near blocks, and that header
can be relayed by SPV peers safely. If the transaction fails to get into
the "seen-but-not-mined" list, a node can use that failure as an
indication to find other peers to relay too.

Currently SPV clients are vulnerable to their peers withholding valid
bloom filter matches; future UTXO commitments can be designed to make
this impossible, and spot-check auditing can detect it now.


Out-of-band fee payment with near blocks
========================================

A purchaser of out-of-band fee payment services can use near-blocks to
check that their fee offer has been accepted and miners are mining their
transaction. This would be particularly useful for a decentralized
system where offers backed by fidelity bonds are made; it would be good
to encourage such systems over arrangements between purchasers and large
pools.


Security
========

There is a serious problem however with proof-of-propagation and
proof-of-mining: they let miners cheat. The proof that a given
transaction is being mined can be used to mine the transaction yourself,
without having to maintain a copy of the UTXO set or indeed do any
validation at all. Having said that this risk already exists due to
P2Pool, which forwards transactions along with shares already.

In any case, it's yet another argument that we need miners to prove they
possess the UTXO set.

-- 
'peter'[:-1]@petertodd.org

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 490 bytes --]