Difference between revisions of "Funding network security"
(Add a page on using network assurance contracts)
Revision as of 13:22, 5 April 2013
One open question is how will funding of network security (mining) work if there's no competition for block space. If funding for proof of work comes from fees attached to transactions and the fees are motivated by scarcity of block space, then the funding mechanism is clear, though whether it will achieve "enough" funding is not.
This page is written by Mike Hearn. Please contact him with questions rather than using the Talk page.
In a world where block sizes are always large enough to meet demand for space, we can fund mining using per-block assurance contracts. From Wikipedia:
- An assurance contract, also known as a provision point mechanism, is a game theoretic mechanism and a financial technology that facilitates the voluntary creation of public goods and club goods in the face of the free rider problem.
- The free rider problem is that there may be actions that would benefit a large group of people, but once the action is taken, there is no way to exclude those who did not pay for the action from the benefits. This leads to a game theoretic problem: all members of a group might be better off if an action were taken, and the members of the group contributed to the cost of the action, but many members of the group may make the perfectly rational decision to let others pay for it, then reap the benefits for free, possibly with the result that no action is taken. The result of this rational game play is lower utility for everyone.
This describes network security with large block sizes. Everyone needs mining to be done, but nobody wants to be the one who pays for it given that everyone else will get the benefits for free.
As described on the contracts wiki page, an assurance contract is one in which an entrepreneur says "I will pay X towards the cost of a public good, but only if other people chip in enough to reach Y (the cost of providing the good)". If not enough people pledge money, the contract does not complete and nobody pays anything. This mechanism has a proven track record of funding public goods, mostly obviously via Kickstarter.
What is the target?
We need to figure out what "enough" mining means. This is related to the maximum time-value of transactions on the network. For micropayments, you don't need a whole lot of mining to protect all of them because the value of reversing them is very low, and any given attacker isn't trying to reverse all transactions but only the one for their own payment (they can't reverse arbitrary transactions). If you have very high value transactions where the receivers are willing to wait six months then you also don't need a whole lot. If you have very high value transactions that are occurring only between people/entities who trust each other, again, don't need much mining. That might sound unrealistic, but once you go above a certain level transactions almost always take place between people who know and can find each other - think about billion dollar business deals, etc. You don't need miners to ensure that won't be reversed. You just need a functioning legal system.
The type of transaction that's most tricky to secure are kind of middle of the road transactions - not billion dollar business deals, not micropayments but the ones where some real value is moving and it's not worth enough to sue the other side if something goes wrong. If there is enough mining to secure this kind of transaction, then there's enough for the other kinds too. There should be a lot of these, and there should also be a lot of participants.
For an assurance contract to work, do you need everyone who benefits to be a participant? No, some freeloading is OK, as long as those freeloaders weren't going to contribute anyway. Let's imagine that 200 major Bitcoin participants get together and form an assurance contract that funds 10 terahashes of mining (numbers are arbitrary). Does their agreement break if 200,000 users then make 500,000 micropayments? Not really - those micropayments hardly need any mining to be secure and those users weren't going to pay for 10 Thash no matter what, not even collectively. There's no downside to them being able to benefit from the extra mining you pay for and indeed, there may be indirect upsides (your Bitcoins become more valuable because they have greater utility).
Implementation can be done effectively via the introduction of a new network rule. It says that money spent to an output that contains an unspendable script can be claimed as fees. We're already introducing the idea of OP_RETURN outputs that simply result in insta-pruning of that output, as it's provably unspendable. Allowing that value to be claimed as fees is a hard-forking rule change, but it's also a relatively straightforward one (the alternative is to have the money be destroyed ... which is bad), and we need to hard fork to increase the block size anyway. Once this is done, we can have a separate P2P broadcast network in which participants broadcast pledges. The pledge is an invalid transaction that takes X coins with a SIGHASH_ANYONECANPAY signature and then re-allocates it to an unspendable output of Y coins, where Y > X. X is the pledge and Y is the target, of course. Peers listen and when they have seen enough pledges to sum up to a value of greater than Y, they just combine them to make the tx valid and broadcast it on the regular Bitcoin network. Peers can do this in parallel - there's a chance of naturally occurring double spends but rules on how to construct the contract out of the pledges can probably reduce that to a trivial level.
Note that it is possible to do spend-to-fee assurance contracts without the new rule, but it'd be complicated and involve a multi-round protocol in which people first announce they want to pledge an output, then someone has to build a template transaction based on all the announcements, then the group signs it in parallel. It can be done but it's messier.
What are X and Y set to? That depends on what the participants want. It'd be nice to find economic research on the case where the public good in question is continuous, but I don't know of any at the moment. I think it'd likely work like this - merchants that have noticed that they start seeing double spends when network speed drops below 5 THash/sec would clearly be targeting that amount and no more. Other merchants might only care about 1 Thash/sec. In that case, the first class of merchants can set their Y to 1 Thash and just broadcast multiple independent contracts, this means there is a chance that they'll get less than what they wanted (which weakens the definition of the target) but there's more of a chance of the contracts completing. At any rate, they would reduce their pledge size as well for each contract, so they aren't exposing themselves to freeloaders at any greater level.
For X, I imagine that if you start from a steady state your goal is to lower it as much as possible without stopping the contracts from completing entirely. One way is to lower your pledge and see how fast the contract completes - if the contract doesn't complete within your required time limit, you can just (anonymously) broadcast a second pledge to make it back to what you were previously pledging. But other players don't know you might do that.
In a healthy system I'd expect there to be many independent, overlapping contracts being formed. They're simple to construct so doing one per block is reasonable.
Obviously whilst there are many participants with different ideas about what network speed is "enough", with only one chain there can only be one speed. People with extreme needs would end up not being able to use the block chain for their security and would have to find alternative arrangements, or just accept that they'd be subsidising the activities of others who can tolerate lower security.
I think the next piece of work needed to explore this idea is searching the literature for studies of assurance contracts for continuous public goods, as the primary difference between what this scheme would do and proven models like Kickstarter is the need for group consensus on what "enough" mining means.
That said, I note that the alternative proposal (restrict block size and charge fees to get in) doesn't even try to answer the question of how much mining is enough. It's just assumed that some arbitrary byte limit would result in demand exceeding supply to the level that mining is well funded. The problem being that limiting blocks by physical size doesn't jive with the fact that they move abstract value - if anything, for such a scheme to work, block sizes should be limited by satoshis transferred. Otherwise you could get a bunch of very high value transactions that end up with tiny fees because that's all that's necessary to get into the block, as lower value transactions have long since given up on the system and gone elsewhere.