This post aims to provide commentary on a paper published last week by Arvind Narayanan & a group of Princeton researchers. I attempt to challenge some of the assumptions presented about the Bitcoin mining incentives system & propose a different interpretation of the analysis provided by the authors.
Fear of the unknown
Once a predetermined number of coins have entered circulation, the incentive can transition entirely to transaction fees and be completely inflation free. — Satoshi Nakamoto, 2008
The transition of Bitcoin’s security subsidy model from block rewards to transaction fees has been a hotly debated topic ever since Bitcoin’s inception.
Uncertainty around the viability of a post block reward environment has spawned many discussions, some formal analysis and various proposals to “fix” what is argued as a situation where the incentives for the miners to secure the network might fall apart.
In their latest paper, Narayanan and his collaborators conclude that fee variance results in perverse incentives that incentivize miners to resort to behaviours harmful to the network in order to try and secure reward for their work.
Fee variance is presented using the “Mining Gap” concept introduced by Miles Carlsten and the same team at the Montreal edition of Scaling Bitcoin.
The argument is rather simple: in the absence of a fixed block reward the potential revenues for miners becomes an unpredictable function of time and, to an extent, the transaction fees available in their memory pool. Considering they incur the costs of burning electricity every second their mining rig is online, the proposition of the authors is that by the time we get there miners may no longer be expected to preserve the current rational mining strategy.
Without a block reward, immediately after a block is found there is zero expected reward for mining but nonzero electricity cost, making it unprofitable for any miner to mine. — Narayanan et al., 2016
The consequences? The paper lays out a number of perverse mining strategies that become incentivized by the “mining gap” and also build on top of previously published ones, such as selfish mining. In short, most of them involve a newly induced potential for miners to make more money by forking the current chain rather than extend it. This obviously creates significant risks with regards to existing Bitcoin security assumptions and potentially undermines the long time sustainability of the system.
How we get around this
The scenario depicted in the paper, from the authors’ point-of-view, is disastrous. Indeed they argue that permanent inflation might just be a necessary evil in the face of such uncertainty.
Perhaps instead, designers of new cryptocurrencies must resign themselves to the inevitability of monetary inflation and make the block reward permanent. — Narayanan et al., 2016
Of course, such a radical redesign of Bitcoin’s monetary policies midway through its existence is a daunting proposition. Suffice to say that given the difficulty observed in the last year in reaching consensus for a block-size limit hard fork, a departure from the initial hard limit of 21 million bitcoins might fracture the ecosystem irreparably.
Fortunately, it happens that there are checks in place in the existing protocol which significantly mitigate the concerns of the authors.
Perhaps the most interesting one, the block size limit, is briefly addressed by the authors:
So if the block size is not large enough to meet demand for transactions, even though the available fees immediately after a block is found will not be zero (as in our analysis), they may be significantly lower than (say) ten minutes later.
For the sake of simplicity Narayanan et al. work within a model that implies miners will always have space available to include whatever transactions are in their mempool. They argue that space scarcity in blocks has little impact on their results given that fee variance still appears to be high in the current environment.
In my opinion, this is the weakest assumption in the paper and one that may have more impact on the real-life application of their findings than they may want to give it credit for.
Today an ordinary node’s mempool holds about 150 blocks worth of transactions, all paying more than 0.00001999 BTC per thousand bytes. A cursory examination of transaction fee dynamics suggest that we are still at the very early stage of a transaction fee market wherein users compete for space in blocks. As transactional demand increases, wallet software improves and Lightning/payment channels settlements become more common it is likely that mempool depth will grow & transaction fees become more uniform. While we cannot extrapolate the extent to which this should happen, it seems equally misguiding to use the current state of the network as a reference.
Additionally, another interesting deterrent against the deviant mining strategies spelled out in the paper is the nLockTime-by-default feature initially proposed by Peter Todd in 2013 and subsequently merged in Bitcoin Core by late 2014. Wallets supporting this will create transactions that can only be processed on top of the current block, severely undermining the chances of successful “fee-sniping”.
After considering all of the above, I cannot help but think of the call to hardfork Bitcoin to a permanent inflationary schedule as an unfortunate sequel to other knee-jerk reactions we’ve seen in the past.
A cautionary tale
While one might disagree with the conclusions of the paper, it does a very good job at spelling out one serious cause of concern that deserves to be considered in the context of the existing scaling debate.
There is a lot of agitation in some camps to increase the quantity of transactions that the network can support on-chain, at all cost. They argue that the block size limit should be kept above market demand with little to no fee pressure. Even more concerning: they would like the miners to be kings and rulers of this market, but that’s one for another day.
I appreciate Narayanan et al.’s paper as empirical observation that an unbounded block size presents significant risk for the network health in the long run.
Without a reasonable expected return miners will necessarily become fee-reliant and in the absence of a hard block size limit the fee variance makes the security of the Bitcoin network unreliable.
We are just barely starting to scratch the surface in terms of improvement and techniques that can be leveraged to make more efficient use of the existing 1 MB limit, let alone SegWit’s expected increase. “Users still don’t care” !
What Narayanan et al. tells us, in my opinion, is that it just might be silly to relieve the pressure now given the long term consequences it implies.
Acknowledgments: I’d like to thank Gregory Maxwell &
for reviewing this post and providing comments.