Very interesting for you to bring this up. I had a similar idea for a totally different use case. Greg recently pointed out an interesting dilemma saying that (significantly) larger blocks would lead to centralization. So I've been working on a design for a decentralized pool that can handle gigabyte sized blocks by splitting up the work among its members.
At the moment P2Pool nodes all verify all transactions in all blocks. But it seems feasible to create a system where miners who have over the last 10000 blocks contributed to the pool's proof-of-work are allocated a proportional piece of verification work with redundancy and deterministic randomness that makes manipulation of the allocation extremely difficult. Such a pool would be very unlikely to accept an invalid block or transaction in practice. However, with these block sizes obviously non miners are going to have to be SPV, so even just a 0.0001% chance of an invalid block being accepted has profound implications for the network. If a decentralized pool like that had more than 50% of the hashing power and it accepted a single invalid block, that tainted chain would be forever regarded as valid by SPV clients. There needs to be some way to recover once an invalid block has been accidentally accepted by an imperfect miner. Based on that I also started to think about proofs of invalidity that would circulate. Basically you would add a new network message that would contain the proof that a specific signature and therefore the whole block is invalid. As long as the block's proof-of-work is valid and the block's parent is one of the last n = 50000 blocks, the message is relayed (subject to a cooldown, warnings would be less frequently relayed the older the offending block is.) The mechanism works in exactly the way Mike mentions: It allows even SPV clients to punish any miner who is dishonest or negligent with their verification work. That gives miners a good reason not to be dishonest or negligent in the first place. (Motivation: Processing more transactions means that hashing is a smaller part of the overall cost for miners. For example, paying for 50 BTC worth of hashing per block costs 0.05 BTC per tx at 1000 tx/block, but only 0.0005 BTC at 100000 tx/block. Number of transactions is a lever that lets us have lower fees and more network security at the same time. Like Greg correctly pointed out, this is not worth having if we have to sacrifice decentralization. But if we don't, it becomes a no-brainer. My IMTUO proposal [1] showed a way where miners don't need a copy of the set of unspent outputs at all. This means the minimum storage requirements per node no longer grow with the number of transactions. However, the price for this was about five times greater bandwidth usage per verified transaction. Since every miner still had to verify every transaction it looked like bandwidth would become an even bigger problem with IMTUO than storage would have been without. However, if a small miner can do less than 100% verifications and still contribute, suddenly IMTUO may become viable. That would accomplish the holy grail of Bitcoin scalability where the network successfully runs on trust-atomic entities all of which can choose to store only a small fraction of the block chain, verify a small fraction of transactions and perform a small fraction of the hashing.) [1] https://en.bitcoin.it/wiki/User:Justmoon/IMTUO On 6/24/2012 2:45 PM, Mike Hearn wrote: > I've been having a discussion with d'aniel from the forums about how > to handle the possibility of a majority-miner conspiracy to raise > inflation, if most economic actors use SPV clients. > > Because of how blocks are formatted you cannot check the coinbase of a > transaction without knowing the fees in the block, and the fees can > only be calculated if you have all the input transactions for every > transaction in that block. Because the attack scenario is an attempted > takeover of the economy by miners, attempting to put hints into the > blocks won't work - we have to assume the hardest chain is in fact > wrong according to the rules signed up to by the Bitcoin user. > > The most obvious goal for a cartel of miners is to change the > inflation formula, either for purely selfish reasons (they want more > money than can be obtained by fees) or due to coercion by > governments/central banks who still subscribe to the "inflation is > good" idea. > > Whilst "good" nodes (still on the old ruleset) won't relay blocks that > violate the rules no matter how hard they are, in a situation where an > SPV client DOES hear about the bad best chain, it would switch to it > automatically. And who knows how the network might look in future - > perhaps most nodes would end up run by miners, or other entities that > upgrade to the new ruleset for other reasons. > > d'aniel made a good proposal - having good nodes broadcast > announcements when they detect a rule that breaks the rules, along > with a proof that it did so. Checking the proof might be very > expensive, but it would only have to be done for split points, > limiting the potential for DoS. If a node announces that it has a > weaker chain and that the split point is a rule-breaker, the SPV > client would download the headers for the side chain to verify the > split, then download all the transactions in the split block along > with all their inputs, and the merkle branches linking the inputs to > the associated block headers. In this way the fee can be calculated, > the inflation formula applied and the coinbase value checked. > > If the block is indeed found to be a rule-breaker, it'd be blacklisted > and chains from that point forward ignored. > > Miners may decide to allow themselves to create money with > non-index-zero transactions to work around this. In that case the good > node can announce that a given tx in the rule-breaker block is > invalid. The SPV node would then challenge nodes announcing the longer > chain to provide the inputs for the bad tx all the way back to a > pre-split coinbase. > > Doing these checks would be rather time consuming with huge blocks, > but it's a last resort only. In the absence of bugs, the mere presence > of the mechanism should ensure it never has to be used. > > ------------------------------------------------------------------------------ > Live Security Virtual Conference > Exclusive live event will cover all the ways today's security and > threat landscape has changed and how IT managers can respond. Discussions > will include endpoint security, mobile security and the latest in malware > threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/ > _______________________________________________ > Bitcoin-development mailing list > Bitcoin-development@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/bitcoin-development ------------------------------------------------------------------------------ Live Security Virtual Conference Exclusive live event will cover all the ways today's security and threat landscape has changed and how IT managers can respond. Discussions will include endpoint security, mobile security and the latest in malware threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/ _______________________________________________ Bitcoin-development mailing list Bitcoin-development@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/bitcoin-development
