On Mon, Nov 04, 2013 at 12:26:30PM +0100, Mike Hearn wrote:
> W.R.T. this paper and the oft-discussed miner backbone,
> 
>   http://arxiv.org/pdf/1311.0243v1.pdf
> 
> I'm wondering about an alternative protocol change that perhaps has less
> subtle implications than their suggested change. Rather than address the
> problem by assuming the network is full of sybil nodes and changing the
> rules for selecting the chain to build on, how about if we wrote code to
> automatically build a miner backbone by having IP addresses of nodes
> embedded into coinbases, then having any bitcoind that is creating work
> automatically connect to IPs that appeared in enough recent blocks?
> 
> It feels like this should be achievable with a few days of solid coding and
> a couple of new command line flags, and the impact is much easier to reason
> about than a fundamental rule change like the one proposed by the paper.

Actually on further reflection this idea will make the attack described
in the paper easier to carry out, rather than harder.

I think where you're misunderstanding originates is the description of
this attack as requiring a sybil attack on the network - you see this
underlying sybil as one of numerical advantage, when it's actually one
of *informational* advantage.

Remember that the selfish miner strategy outlined in the paper is
essentially a way to use knowledge of what blocks miners will be mining
on, from the "first seen" rule, and the ability to broadcast blocks you
have mined more widely than other miners. That knowledge and ability is
then used in conjunction with a small lead (obtainable by chance) to
outpace the rest of the network.

By making all miners easily identifiable you make gaining that
informational and broadcast capability easier to obtain rather than
harder. The attacker now only needs to connect to every identified miner
with especially fast nodes. With judicious use of DoS attacks and low
latency they can still gain the informational and broadcast "upper hand"
over other miners and carry out the attack.

Where the paper goes wrong is they don't recognize the fundemental
nature of the strategy being based on an informational advantage. Their
"pick a random side of the fork" strategy may work to some extent, but
it's incomplete and isn't necessarily rational for the miners
individually.

The correct, and rational, approach for a miner is to always mine to
extend the block that the majority of hashing power is trying to extend.
The current relay rules don't give you that information at all, but they
can if we do two things:

1) Relay all blocks that meet the PoW target. (as suggested in the
   paper)

2) Relay block headers that nearly meet the PoW target.

Mining strategy is now to mine to extend the first block you see, on the
assumption that the earlier one probably propagated to a large portion
of the total hashing power. But as you receive "near-blocks" that are
under the PoW target, use them to estimate the hashing power on each
fork, and if it looks like you are not on the majority side, switch.

This very effectively defeats the paper's selfish-miner strategy, as all
miners will very quickly be mining on the block that truly has the
majority of hashing power trying to extend it. This is also a better
overall outcome, because it puts the 51% attack threshhold back at 51%

-- 
'peter'[:-1]@petertodd.org
0000000000000004ee9bb13b022c412d75692b5e85454013c53f89e5d6fa8c69