MIME-Version: 1.0
From: Larry Ruane <larryruane@gmail.com>
Date: Tue, 22 Mar 2022 13:04:26 -0600
Message-ID: <CAEpYn+eOv7xRkTA2RfpC+ps2ykyvjfZ-iY8-4z9nHov-dNLimw@mail.gmail.com>
To: Bitcoin Protocol Discussion <bitcoin-dev@lists.linuxfoundation.org>
Content-Type: text/plain; charset="UTF-8"
Subject: [bitcoin-dev] mempool transaction witness-replacement
Precedence: list

Greetings list,

This is my first time posting here.

Question for you:

Should the Bitcoin Core mempool replace an existing transaction with one
that has the same txid (having the same effect, same spends and outputs)
but a sufficiently smaller witness (different wtxid) and thus a higher
feerate? This is what https://github.com/bitcoin/bitcoin/pull/24007
proposes, and I'd like to get opinions on two questions:

1. Is this a beneficial change? Specifically, is anyone creating an
application that would broadcast transactions with the same txid but
different witnesses as an earlier transaction?

2. If this change has benefit, what should be considered a sufficiently
better feerate or reduction in witness size?

An advantage of this mempool-accept policy change is that it's
miner-incentive compatible (miners would prefer to mine a transaction
with a higher feerate). But there is of course a code complexity cost,
and transaction-relay DoS concern.

Being miner-incentive compatible is good, but is that sufficient
justification for merging? I'm posting to the mailing list in hopes that
there are use-cases that we (the PR authors) aren't aware of. Please
reply here or on the PR if you can think of any.

A perhaps subtle advantage: This PR may provide a defense against a
mempool pinning attack: if you have a transaction shared with other
parties, and one of them broadcasts the transaction with a bloated
witness (thus intentionally reducing the feerate in hopes of delaying
or preventing confirmation), you currently have no way to change it.
If there is an application out there that uses same-txid-different-witness
transactions shared between counterparties, this PR would help make
those applications safe.

Question 2 gets at a DoS tradeoff: If the new transaction may have
only a very slightly smaller witness, an attacker might re-broadcast it
many times, consuming a lot of relay bandwidth, and CPU to update
the mempool. On the other hand, if the new transaction must have a much
smaller witness, then it wouldn't be possible to replace a transaction with
a beneficially-smaller one.

This could be a per-node setting, but it's desirable for the node
network to largely agree on relay policies (although a configuration
option might be useful for testing and experimentation).

Background:

Bip125 (Replace-by-fee) allows an incoming transaction to replace one
or more existing conflicting transactions if certain DoS-mitigation
conditions are met:

https://github.com/bitcoin/bitcoin/blob/master/doc/policy/mempool-replacements.md

Witness-replacement is similar to RBF, but differs in a few ways:

- RBF rule 4 requires an absolute fee increase, which is not possible if
the txid isn't changing (since the inputs, outputs, and amounts must be
the same). So if transaction witness-replacement (same txid but different
wtxid) is allowed, it can't be considered just a special case of an RBF,
although it may have some similar policies (and for the same reasons).

- With witness-replacement, it's not necessary to evict mempool
descendant transactions because their inputs' txid references to their
parent (who is being replaced) remain valid.

- The new transaction replaces exactly one existing transaction since
the inputs are the same. (In general, with RBF, the new transaction may
conflict-out multiple existing mempool transactions, namely, all that
spend the same outputs as the new transaction.)

- RBF requires the original transaction to signal replaceability
(rule 1). This is so that recipients are warned that their payment may
disappear if the transaction is replaced. But signaling isn't required
by witness-replacement since the outputs can't change (the descendants
remain valid).

Thanks for your time!

Larry Ruane (with lots of help from Gloria Zhao)