Date: Thu, 3 Oct 2019 22:20:31 -1000
From: "David A. Harding" <dave@dtrt.org>
To: Braydon Fuller <braydon@purse.io>,
	Bitcoin Protocol Discussion <bitcoin-dev@lists.linuxfoundation.org>
Message-ID: <20191004082031.ns3pgzwh2zz2mxyc@ganymede>
References: <42cd5ffd-63e8-b738-c4ea-13d0699b1268@purse.io>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <42cd5ffd-63e8-b738-c4ea-13d0699b1268@purse.io>
User-Agent: NeoMutt/20180716
Subject: Re: [bitcoin-dev] Chain width expansion
Precedence: list

On Thu, Oct 03, 2019 at 05:38:36PM -0700, Braydon Fuller via bitcoin-dev wrote:
> This paper describes a solution [to DoS attacks] that does not
> require enabling or maintaining checkpoints and provides improved security.
> [...] 
> The paper is available at:
> https://bcoin.io/papers/bitcoin-chain-expansion.pdf

Hi Braydon,

Thank you for researching this important issue.  An alternative solution
proposed some time ago (I believe originally by Gregory Maxwell) was a
soft fork to raise the minimum difficulty.  You can find discussion of
it in various old IRC conversations[1,2] as well as in related changes
to Bitcoin Core such as PR #9053 addining minimum chain work[3] and the
assumed-valid change added in Bitcoin Core 0.14.0[4].

[1] http://www.erisian.com.au/meetbot/bitcoin-core-dev/2016/bitcoin-core-dev.2016-10-27-19.01.log.html#l-121
[2] http://www.erisian.com.au/meetbot/bitcoin-core-dev/2017/bitcoin-core-dev.2017-03-02-19.01.log.html#l-57
[3] https://github.com/bitcoin/bitcoin/pull/9053/commits/fd46136dfaf68a7046cf7b8693824d73ac6b1caf
[4] https://bitcoincore.org/en/2017/03/08/release-0.14.0/#assumed-valid-blocks

The solutions proposed in section 4.2 and 4.3 of your paper have the
advantage of not requiring any consensus changes.  However, I find it
hard to analyze the full consequences of the throttling solution in
4.3 and the pruning solution in 4.2.  If we assume a node is on the
most-PoW valid chain and that a huge fork is unlikely, it seems fine.
But I worry that the mechanisms could also be used to keep a node that
synced to a long-but-lower-PoW chain on that false chain (or other false
chain) indefinitely even if it had connections to honest peers that
tried to tell it about the most-PoW chain.

For example, with your maximum throttle of 5 seconds between
`getheaders` requests and the `headers` P2P message maximum of 2,000
headers per instance, it would take about half an hour to get a full
chain worth of headers.  If a peer was disconnected before sending
enough headers to establish they were on the most-PoW chain, your
pruning solution would delete whatever progress was made, forcing the
next peer to start from genesis and taking them at least half an hour
too.  On frequently-suspended laptops or poor connections, it's possible
a node could be be operational for a long time before it kept the same
connection open for half an hour.  All that time, it would be on a
dishonest chain.

By comparison, I find it easy to analyze the effect of raising the
minimum difficulty.  It is a change to the consensus rules, so it's
something we should be careful about, but it's the kind of
basically-one-line change that I expect should be easy for a large
number of people to review directly.  Assuming the choice of a new
minimum (and what point in the chain to use it) is sane, I think it
would be easy to get acceptance, and I think it would further be easy
increase it again every five years or so as overall hashrate increases.

-Dave