On 05/29/15 23:48, Gavin Andresen wrote:
On Fri, May 29, 2015 at 7:25 PM, Matt Corallo bitcoin-l...@bluematt.me
mailto:bitcoin-l...@bluematt.me wrote:
Sadly, this is very far from the whole story. The issue of miners
optimizing for returns has been discussed several times during this
discussion, and, sadly, miners who are geographically colocated who are
optimizing for returns with a free-floating blocksize will optimize away
50% of the network!
I must have missed that analysis-- link please? Or summary of HOW they
will optimize away 50% of the network?
Or are you assuming that 50% of the network is colocated... (which is a
potential problem independent of blocksize)
If, for example, the majority of miners are in China (they are), and
there is really poor connectivity in and out of China (there is) and a
miner naively optimizes for profit, they will create blocks which are
large and take a while to relay out of China. By simple trial-and-error
an individual large miner might notice that when they create larger
blocks which fork off miners in other parts of the world, they get more
income. Obviously forking off 50% of the network would be a rather
extreme situation and assumes all kinds of simplified models, but it
shows that the incentives here are very far from aligned, and your
simplified good-behavior models are very far from convincing.
In addition, I'd expect to
see analysis of how these systems perform in the worst-case, not
just
packet-loss-wise, but in the face of miners attempting to break the
system.
See
http://gavinandresen.ninja/are-bigger-blocks-better-for-bigger-miners for
analysis of but that means bigger miners can get an advantage
argument.
Executive summary: if little miners are stupid and produce huge blocks,
then yes, big miners have an advantage.
I'll talk about transaction fees in a second, but there are several
problems with this already. As pointed out in the original mail, gfw has
already been known to interfere with Bitcoin P2P traffic. So now by
little miners, you mean any miner who is not located in mainland
China? Whats worse, the disadvantage is symmetric - little miners are at
a disadvantage when *anyone* mines a bigger block, and miners dont even
have to be evil for this to happen - just optimize for profits.
But the disadvantage is tiny. And essentially zero if they connect to
your fast relay network (or anything like it).
The disadvantage is small with 1MB blocks, but already non-zero. 20MB
blocks are much, much worse (lots of things here dont scale linearly,
even just transfer over a high-packet-loss-link). I mentioned this in my
original email as something which doesnt make me comfortable with 20MB
blocks, but something which needs simulation and study, and might
actually be just fine!
But they're not, so they won't.
I dont know what you're referring to with this. Are you claiming little
miners today optimize for relay times and have good visibility into the
Bitcoin network and calculate an optimal block size based on this (or
would with a 20MB block size)?
Do you have another explanation for why miners choose to leave
fee-paying transactions in their mempool and create small blocks?
Defaults? Dumb designs? Most miners just use the default 750K blocks, as
far as I can tell, other miners probably didnt see transactions relayed
across several hops or so, and a select few miners are doing crazy
things like making their blocks fit in a single packet to cross the gfw,
but that is probably overkill and not well-researched.
Until the block reward goes away, and assuming transaction fees become
an important source of revenue for miners.
I think it is too early to worry about that; see:
http://gavinandresen.ninja/when-the-block-reward-goes-away
You dont make any points here with which I can argue, but let me respond
with the reason /I/ think it is a problem worth thinking a little bit
about...If we increase the blocksize sufficiently such that transaction
fees are not the way in which miners make their money
I'm not suggesting that we increase the blocksize sufficiently such that
transaction fees are not the way in which miners make their money.
I'm suggesting the blocksize be increased to 20MB (and then doubled
every couple of years).
Do you have convincing evidence that at 20MB miners will be able to
break even on transaction fees for a long time? (The answer is no
because no one has any idea how bitcoin transaction volumes are going to
scale, period.)
And in which miners make their money is the wrong metric-- we want
enough mining so the network to be secure enough against double-spends.
Sure, do you have a value of hashpower which is secure enough (which
is a whole other rabbit hole to