This is the sixth in a series analyzing the pettycoin implementation against Gregory Maxwell’s writeup on scaling. The first talked about UTXO commitments vs backrefs, the second talked about Propogation servers vs prev_txhashes, the third talked about Hashed Fees vs Random Extrapolation, the fourth talked about Simplified Transactions vs Normal Bitcoin Transactions, and the fifth discussed Fast blocks vs Bitcoin blocks.
Mining fees must be sufficient to support the network, otherwise it will eventually fail. There’s a tension between users who want the lowest fees possible, and miners who want the highest fees possible. And while the miner of a particular block collects the fees, the incremental cost of remembering the transaction is carried by the entire network.
Bitcoin is supposed to use an open market fee structure; miners choose what to accept based on fees, and so transactions offer a fee sufficient to have miners include the transaction in their block.
In practice, the default formula used is very simple: there are a number of “first-in-first-served” free transaction slots in each block, then a fixed fee is required per kb of transaction. This fixed fee has dropped several times as the bitcoin price increases.
It’s not particularly important, as miners view transactions fees as a tiny cherry on top: the block rewards are driving mining at the moment. This will eventually change, so efforts are underway to have the reference bitcoin client guestimate appropriate fees which are likely to have a transaction included. How well this will work in practice is uncertain: miners have short-term incentive to include transactions no matter how small the fee, but a longer-term incentive to reject tiny fees. This is especially true if clients autodetect appropriate fee levels.
As a first approximation, the cost of running the bitcoin network should be equal to the value of bitcoins currently produced. Let’s assume that we want the network to continue at the current spending level when the block reward next halves; transaction fees will have to make up the difference, at around $4000 per block.
The bitcoin price is a huge unknown, so we’ll try to make our other guesses extremely conservative. Let the bitcoin transaction rate will increase to a massive 100 transactions per second, that’s 60,000 transactions each paying around 6c each.
If bitcoin won’t let you reasonably send 1c, what would? A sidechain specifically designed for this, with properties that ensured it wouldn’t compete with bitcoin. That way, miners could pick up the (too-small) fees, without entering a race to the bottom on the bitcoin network.
Of course, the sidechain would have to be cheap to run, too, so the additional cost of merge mining it was worth the small fees involved.
This was the core design of pettycoin:
The horizon ensures that pettycoin is not a store of value, but only a transfer network, and puts an approximate upper limit on how much data needs to be stored.
Simplified transactions were a bad idea as they don’t save much, and it’s hard to tell which parts of the bitcoin scripts will be useful in future.
The percentage-based fee (0.3% currently) makes for a simple penalty if transactions get large enough that it would be cheaper to do them on the bitcoin network. (You could also flag a transaction as zero-fee, with the idea that miners may or may not accept them).
This scheme required simplified transactions, which can distinguish a payment from change. Using percentage of the total transferred would encourage users to split their outputs so future transactions would not need to transfer as much, leading to a larger UTXO size.
The simplest solution I can think of would be to ignore the largest output, and sum the rest as the fee basis.
This could still be gamed for some cases, of course, but is fairly simple in the common case. There’s no reason to restrict number of inputs, as consuming many inputs would reduce the UTXO size and help the network overall.
This core idea of a complementary network for sub-cent transactions has promise, but has a significant bootstrap problem. If there are still free bitcoin transactions, it provides no value. If there are no more free bitcoin transactions, then you need to pay the transaction fee to get money on and off the sidechain, in order to save money later (and you only have a limited time to do it).
It would take a compelling use case to drive adoption, which could be artificial: eg. a subsidised SatoshiDice-style service on the microtransaction network.