MIME-Version: 1.0
References: <9Vw6LCkr2d2uOBanXeIuGxA1fUGGOeV1OHlgBifbmij1Afs0ISjfKK-vmcnRZfBG4GwJhIVLMisjvS_zohS-cW0FkzZaCKa6Mn7VWolznJs=@protonmail.com>
 <8be86b19-04eb-af12-a54c-e1140ac62e3f@mattcorallo.com>
 <20220217143225.GB1429@erisian.com.au>
In-Reply-To: <20220217143225.GB1429@erisian.com.au>
From: "James O'Beirne" <james.obeirne@gmail.com>
Date: Thu, 17 Feb 2022 13:18:11 -0500
Message-ID: <CAPfvXfLRBdk5-pgX3z+azO+z5DAiGhcR+JRWizhXw9TcYnPhsw@mail.gmail.com>
To: Anthony Towns <aj@erisian.com.au>
Content-Type: multipart/alternative; boundary="0000000000009e7c2b05d83acab7"
Cc: Bitcoin Protocol Discussion <bitcoin-dev@lists.linuxfoundation.org>
Subject: Re: [bitcoin-dev] Thoughts on fee bumping
Precedence: list

--0000000000009e7c2b05d83acab7
Content-Type: text/plain; charset="UTF-8"

> Is it really true that miners do/should care about that?

De facto, any miner running an unmodified version of bitcoind doesn't
care about anything aside from ancestor fee rate, given that the
BlockAssembler as-written orders transactions for inclusion by
descending ancestor fee-rate and then greedily adds them to the block
template. [0]

If anyone has any indication that there are miners running forks of
bitcoind that change this behavior, I'd be curious to know it.

Along the lines of what AJ wrote, optimal transaction selection is
NP-hard (knapsack problem). Any time that a miner spends deciding how
to assemble the next block is time not spent grinding on the nonce, and
so I'm skeptical that miners in practice are currently doing anything
that isn't fast and simple like the default implementation: sorting
fee-rate in descending order and then greedily packing.

But it would be interesting to hear evidence to the contrary.

---

You can make the argument that transaction selection is just a function
of mempool contents, and so mempool maintenance criteria might be the
thing to look at. Mempool acceptance is gated based on a minimum
feerate[1].  Mempool eviction (when running low on space) happens on
the basis of max(self_feerate, descendant_feerate) [2]. So even in the
mempool we're still talking in terms of fee rates, not absolute fees.

That presents us with the "is/ought" problem: just because the mempool
*is* currently gating only on fee rate doesn't mean that's optimal. But
if the whole point of the mempool is to hold transactions that will be
mined, and if there's good reason that txns are chosen for mining based
on fee rate (it's quick and good enough), then it seems like fee rate
is the approximation that should ultimately prevail for txn
replacement.


[0]:
https://github.com/bitcoin/bitcoin/blob/master/src/node/miner.cpp#L310-L320
[1]:
https://github.com/bitcoin/bitcoin/blob/master/src/txmempool.cpp#L1106
[2]:
https://github.com/bitcoin/bitcoin/blob/master/src/txmempool.cpp#L1138-L1144

--0000000000009e7c2b05d83acab7
Content-Type: text/html; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

<div dir=3D"ltr">&gt; Is it really true that miners do/should care about th=
at?<br><br>De facto, any miner running an unmodified version of bitcoind do=
esn&#39;t<br>care about anything aside from ancestor fee rate, given that t=
he<br>BlockAssembler as-written orders transactions for inclusion by<br>des=
cending ancestor fee-rate and then greedily adds them to the block<br>templ=
ate. [0]<br><br>If anyone has any indication that there are miners running =
forks of<br>bitcoind that change this behavior, I&#39;d be curious to know =
it.<br><br>Along the lines of what AJ wrote, optimal transaction selection =
is<br>NP-hard (knapsack problem). Any time that a miner spends deciding how=
<br>to assemble the next block is time not spent grinding on the nonce, and=
<br>so I&#39;m skeptical that miners in practice are currently doing anythi=
ng<br>that isn&#39;t fast and simple like the default implementation: sorti=
ng<br>fee-rate in descending order and then greedily packing. <br><br>But i=
t would be interesting to hear evidence to the contrary.<br><br>---<br><br>=
You can make the argument that transaction selection is just a function<br>=
of mempool contents, and so mempool maintenance criteria might be the<br>th=
ing to look at. Mempool acceptance is gated based on a minimum<br>feerate[1=
].=C2=A0 Mempool eviction (when running low on space) happens on<br>the bas=
is of max(self_feerate, descendant_feerate) [2]. So even in the<br>mempool =
we&#39;re still talking in terms of fee rates, not absolute fees. <br><br>T=
hat presents us with the &quot;is/ought&quot; problem: just because the mem=
pool<br>*is* currently gating only on fee rate doesn&#39;t mean that&#39;s =
optimal. But<br>if the whole point of the mempool is to hold transactions t=
hat will be<br>mined, and if there&#39;s good reason that txns are chosen f=
or mining based<br>on fee rate (it&#39;s quick and good enough), then it se=
ems like fee rate<br>is the approximation that should ultimately prevail fo=
r txn<br>replacement.<br><br><br>[0]:<br><a href=3D"https://github.com/bitc=
oin/bitcoin/blob/master/src/node/miner.cpp#L310-L320">https://github.com/bi=
tcoin/bitcoin/blob/master/src/node/miner.cpp#L310-L320</a><br>[1]:<br><a hr=
ef=3D"https://github.com/bitcoin/bitcoin/blob/master/src/txmempool.cpp#L110=
6">https://github.com/bitcoin/bitcoin/blob/master/src/txmempool.cpp#L1106</=
a><br>[2]:<br><a href=3D"https://github.com/bitcoin/bitcoin/blob/master/src=
/txmempool.cpp#L1138-L1144">https://github.com/bitcoin/bitcoin/blob/master/=
src/txmempool.cpp#L1138-L1144</a><br></div>

--0000000000009e7c2b05d83acab7--