Bitcoin: ASICBoost and segwit2x – Background

I’ve been trying to make heads or tails of what the heck is going on in Bitcoin for a while now. I’m not sure I’ve actually made that much progress, but I’ve at least got some thoughts that seem coherent now.

First, this post is background for people playing along at home who aren’t familiar with the issues or jargon: Bitcoin is a currency based on an electronic ledger that essentially tracks how much Bitcoin exists, and how someone can be authorised to transfer it to someone else; that ledger is currently about 100GB in size, growing at a rate of about a gigabyte a week. The ledger is updated by miners, who compete by doing otherwise pointless work running cryptographic hashes (and in so doing obtain a “proof of work”), and in return receive a reward (denominated in bitcoin) made up from fees by people transacting and an inflation subsidy. Different miners are competing in an essentially zero-sum game, because fees and inflation are essentially a fixed amount that is (roughly) divided up amongst miners according to how much work they do — so while you get more reward for doing more work, it comes at a cost of other miners receiving less reward.

Because the ledger only grows by (about) a gigabyte each week (or a megabyte per block, which is roughly every ten minutes), there is a limit on how many transactions can be included each week (ie, supply is limited), which both increases fees and limits adoption — so for quite a while now, people in the bitcoin ecosystem with a focus on growth have wanted to work out ways to increase the transaction rate. Initial proposals in mid 2015 suggested allowing miners to regularly determine the limit with no official upper bound (nominally “BIP100“, though never actually formally submitted as a proposal), or to increase by a factor of eight within six months, then double every two years after that, until reaching almost 200 times the current size by 2036 (BIP101), or to increase at a rate of about 17% per annum (suggested on the mailing list, but never formally proposed BIP103). These proposals had two major risks: locking in a lot of growth that may turn out to be unnecessary or actively harmful, and requiring what is called a “hard fork”, which would render the existing bitcoin software unable to track the ledger after the change took affect with the possible outcome that two ledgers would coexist and would in turn cause a range of problems. To reduce the former risk, a minimal compromise proposal was made to “kick the can down the road” and just double the ledger growth rate, then figure out a more permanent solution down the road (BIP102) (or to double it three times — to 2MB, 4MB then 8MB — over four years, per Adam Back). A few months later, some of the devs figured out a way to more or less achieve this that also doesn’t require a hard fork, and comes with a host of other benefits, and proposed an update called “segregated witness” at the December 2015 Scaling Bitcoin conference.

And shortly after that things went completely off the rails, and have stayed that way since. Ultimately there seem to be two camps: one group is happy to deploy segregated witness, and is eager to make further improvements to Bitcoin based on that (this is my take on events); while the other group does not, perhaps due to some combination of being opposed to the segregated witness changes directly, wanting a more direct upgrade immediately, being afraid deploying segregated witness will block other changes, or wanting to take control of the bitcoin codebase/roadmap from the current developers (take this with a grain of salt: these aren’t opinions I share or even find particularly reasonable, so I can’t do them justice when describing them; cf ViaBTC’s post to get that side of the argument made directly, eg)

Most recently, and presumably on the basis that the opposed group are mostly worried that deploying segregated witness will prevent or significantly delay a more direct increase in capacity, a bitcoin venture capitalist, Barry Silbert, organised an agreement amongst a number of companies including many miners, to both activate segregated witness within the next month, and to do a hard fork capacity increase by the end of the year. This is the “segwit2x” project; named because it takes segregated witness, (“segwit”) and then additionally doubles its capacity increase (“2x”). This agreement is not supported by any of the existing dev team, and is being developed by Jeff Garzik (who was behind BIP100 and BIP102 mentioned above) in a forked codebase renamed “btc1“, so if successful, this may also satisfy members of the opposed group motivated by a desire to take control of the bitcoin codebase and roadmap, despite that not being an explicit part of the agreement itself.

To me, the arguments presented for opposing segwit don’t really seem plausible. As far as future development goes, a roadmap was put out in December 2015 and endorsed by many developers that explicitly included a hard fork for increased capacity (“moderate block size increase proposals (such as 2/4/8 …)”), among many other things, so the risk of no further progress happening seems contrary to the facts to me. The core bitcoin devs are extremely capable in my estimation, so replacing them seems a bad idea from the start, but even more than that, they take a notably hands off approach to dictating where Bitcoin will go in future — so, to my mind, it seems like a more sensible thing to try would be working with them to advance the bitcoin ecosystem in whatever direction you want, rather than to try to replace them outright. In that context, it seems particularly notable to me that in the eighteen months between the segregated witness proposal and the segwit2x agreement, there hasn’t been any serious attempt to propose a hard fork capacity increase that meets the core dev’s quality standards; for instance there has never been any code for BIP100, and of the various hard forking codebases that have arisen by advocates of the hard fork approach — Bitcoin XT, Bitcoin Classic, Bitcoin Unlimited, btc1, and Bitcoin ABC — none have been developed in a way that’s suitable for the changes to be reviewed and merged into core via a pull request in the normal fashion. Further, since one of the main criticisms of a hard fork is that deployment costs are higher when it is done in a short time frame (weeks or a few months versus a year or longer), that lack of engagement over the past 18 months followed by a desperate rush now seems particularly poor to me.

A different explanation for the opposition to segwit became public in April, however. ASICBoost is a patent-pending optimisation to the way Bitcoin miners do the work that entitles them to extend the ledger (for which they receive the rewards described earlier), and while there are a few ways of making use of ASICBoost, perhaps the most effective way turns out to be incompatible with segwit. There are three main alternatives to the covert, segwit-incompatible approach, all of which have serious downsides. The first, overt ASICBoost via modifying the block version reveals that you’re using ASICBoost, which would either (a) encourage other miners to also use the optimisation reducing your profits, (b) give the patent holder cause to charge you royalties or cause other problems (assuming the patent is eventually granted and deemed valid), or (c) encourage the bitcoin community at large to change the ecosystem rules so that the optimisation no longer works. The second, mining empty blocks via ASICBoost means you don’t gain any fee income, reducing your revenue and hence profit. And the third, rolling the extranonce to find a collision rather than combining partial transaction trees increases the preparation work by a factor of ten or so, which is probably enough to outweigh the savings from the optimisation in the first place.

If ASICBoost were being used by a significant number of miners, and segregated witness prevents its continued use in practice, then we suddenly have a very plausible explanation for much of the apparent madness: the loss of the optimisation could significantly increase some miners’ costs or reduce their revenue, reducing profit either way (a high end estimate of $100,000,000 per year was given in the original explanation), which would justify significant investment in blocking that change. Further, an honest explanation of the problem would not be feasible, because this would be just as bad as doing the optimisation overtly — it would increase competition, alert the potential patent owners, and might cause the optimisation to be deliberately disabled — all of which would also negatively affect profits. As a result, there would be substantial opposition to segwit, but the reasons presented in public for this opposition would be false, and it would not be surprising if the people presenting these reasons only give half-hearted effort into providing evidence — their purpose is simply to prevent or at least delay segwit, rather than to actually inform or build a new consensus. To this line of thinking the emphasis on lack of communication from core devs or the desire for a hard fork block size increase aren’t the actual goal, so the lack of effort being put into resolving them over the past 18 months from the people complaining about them is no longer surprising.

With that background, I think there are two important questions remaining:

  1. Is it plausible that preventing ASICBoost would actually cost people millions in profit, or is that just an intriguing hypothetical that doesn’t turn out to have much to do with reality?
  2. If preserving ASICBoost is a plausible motivation, what will happen with segwit2x, given that by enabling segregated witness, it does nothing to preserve ASICBoost?

Well, stay tuned…

Leave a Reply