On Tue, Jun 23, 2015 at 05:24:23PM -0400, Gavin Andresen wrote: > On Tue, Jun 23, 2015 at 4:46 PM, Peter Todd wrote: > > > Pieter Wuille showed with simulations that miners with bad connectivity > > are negatively affected by other miners creating larger blocks. > > > > ... but the effect is only significant if they have an absurdly > low-bandwidth connection and do NOTHING to work around it (like rent a > server on the other side of the bandwidth bottleneck and write some code to > make sure you're creating blocks that will propagate quickly on both sides > of the bottleneck). "Just rent a server" forces miners into deploying insecure hosted infrastructure that's vulnerable to hacking and seizure; that we encourage this already is worrying; requiring it for miners to be profitable isn't acceptable. > Why do you think connectivity is a centralizing effect? It is just one > factor in the profitability-of-mining equation. A location with bad > connectivity (the US, maybe) but 10% cheaper electricity might be just as > good as one with great connectivity but more expensive electricity. See above. The obvious thing to do if connectivity matters is keep your hashing in the cheapest possible place and sell that hashing power to centralized miners, an effect we're already seeing. Making this effect about an order of magnitude worse, then doubling the problem every two years is dangerous. > Having lots of variables in the profitability equation is a decentralizing > force, it means there is very likely to be several different places in the > world / on the net where mining is equally profitable. As mining and hashing can be trivially separated that theory just doesn't work. Again, what concretely works against centralization of mining control? A proper proposal would discuss this issue, and explain what the expected effect will be. > > These block propagation improvements are both already implemented (Matt > > Corallo's relay network, p2pool) and require co-operation. > > > > Long term the p2p protocol will evolve to incorporate those optimizations, > so will require no co-operation. The co-operation comes form the fact that mempool policies have to be syncronized, not the protocol itself. -- 'peter'[:-1]@petertodd.org 0000000000000000007fc13ce02072d9cb2a6d51fae41fefcde7b3b283803d24