Dear Jameson,

Thank you for your questions. Answers inline below:

Jameson Lopp via bitcoin-dev <bitcoin-dev@lists.linuxfoundation.org> wrote:
You've made many salient points, Shaolin, though I have a few questions:

1) How well does this model work under adversarial conditions? Fair point about signaling not being reliable, though it seems more vague in terms of safety given that you can't actually know what percentage of hashrate that is /not/ signaling for the soft fork has taken the necessary precautions to avoid mining an invalid block and potentially causing a hard fork. It's probably safe to say that if a flag-day soft fork is activated, there will be at least a few parties who will attempt to trigger a chain fork by crafting transactions that are valid via non-fork rules but invalid via the soft fork rules.

In a well designed soft fork, transactions under the old rules are non-standard by default and will not propagate or be mined. A miner would have to deliberately include the invalid transaction in a block and mine it. The invalid block would be rejected by the network costing the miner block reward and fees.

If >51% of the hashrate does not upgrade or does not take steps to protect themselves from invalid blocks, they will fork if someone produces an invalid block. Game theory suggests the incentive for those who do not wish to participate, would be to do so safely. There is no incentive to allow an attacker to cause you to split off from the network and it is trivial to prevent it.

There is a valid concern about "spy" mining and I cited a previous incident with BIP66 activation and we should be working towards solutions that remove the incentive to spy mine. "Weak blocks", where miners propagate their proposed blocks before solving the PoW may provide better incentives against spy mining, while delivering more (~no propagation delay and full validation, and thus more security).

2) If the flag day soft fork is activated with only a minority of hashrate support + safely opted-out hashrate, isn't it possible for the rest of miners to coordinate orphaning any soft fork compatible blocks to kill the soft fork chain? This would be a major difference from a miner-activated soft fork, correct? Unless perhaps many miners colluded to signal soft fork support while not actually supporting it...

The basic assumption in the Bitcoin system is that miners will remain honest because it is in their economic interest to do so. Of course 51% of the hashrate can censor the minority hash by orphaning blacklisted transactions or blocks. I am fairly certain it would be considered an attack by as well as being very conspicuous. A 51% attack would likely cause a dramatic loss in confidence in the Bitcoin system and adversely affect price. It is reasonable to assume miners would not do that because mining has to remain profitable. Additionally, such a scenario would draw much ire from users who may escalate demands for a PoW change.

It is assuming good-faith and that miners would not want to deny people the ability to opt into something they wanted. All that is required of miners is to upgrade their border node. Miners should update their software anyway for security reasons.

3) In terms of complexity for mining pool operators, how well does this model scale if there are N soft forks and the pool doesn't want to opt-in to any of them? Couldn't this result in those pool operators having to run not just one border node, but a multitude of "chained" border nodes if the soft forks are spread across different software implementations?

While BIP9 allows for 29 parallel deployments I think it is unrealistic to expect there would be such a high number of active parallel deployments at any one time: History shows soft forks take a minimum of 6 months design, consensus building, coding and testing before deployment. With such a high bar, I do not envisage more than a couple of parallel deployments at any given time. I also do not envisage "conflicting" soft forks, as that would not meet consensus from the technical community on the basis of safety and sanity. In any case, the deployment strategy of each soft fork should be considered on a case by case basis.

It seems to me that this type of user-driven approach would preferably be coupled with assurances from major Bitcoin wallets / exchanges / payment processors that they will not honor coins from a chain fork that results from invalid spends of outputs encumbered by soft fork rules. Though on the other hand, I don't see such an assurance being possible given that exchanges have an incentive to take the first mover advantage in listing a new coin.

Soft fork consensus proposals should be sane, uncontroversial and have a reasonably high bar in terms of technical consensus as we have seen with other soft forks to date. There is an implicit assumption in my text, that the decision to deploy a soft fork (regardless of the activation method) is based on a reasonable expectation that users will make use of the new feature. Hashrate signalling is not a vote, but a coordination trigger. Soft forks are backwards compatible and opt-in; so long as they are well written and bug free, users should at worst, be agnostic towards them because they have a choice whether to safely use the new feature or not, without preventing others' enjoyment of the feature. A controversial or unreasonable soft fork would not gain traction and I believe it would be fairly self evident.

In short, I do expect wide ecosystem collaboration as part of any deployment strategy, both hashrate or flag day based.

Many thanks for taking the time to read over and consider my thoughts and proposal. I would be happy to discuss more if you have any further questions or suggestions.

- Jameson

On Sat, Feb 25, 2017 at 6:55 PM, shaolinfry via bitcoin-dev <bitcoin-dev@lists.linuxfoundation.org> wrote:
Some thoughts about the activation mechanism for soft forks. In the past we used IsSuperMajority and currently use BIP9 as soft fork activation methods, where a supermajority of hashrate triggers nodes to begin enforcing new rules. Hashrate based activation is convenient because it is the simplest and most straightforward process. While convenient there are a number limitations with this method.

Firstly, it requires trusting the hash power will validate after activation. The BIP66 soft fork was a case where 95% of the hashrate was signaling readiness but in reality about half was not actually validating the upgraded rules and mined upon an invalid block by mistake[1].

Secondly, miner signalling has a natural veto which allows a small percentage of hashrate to veto node activation of the upgrade for everyone. To date, soft forks have taken advantage of the relatively centralised mining landscape where there are relatively few mining pools building valid blocks; as we move towards more hashrate decentralization, it's likely that we will suffer more and more from "upgrade inertia" which will veto most upgrades.

Upgrade inertia in inevitable for widely deployed software and can be seen for example, with Microsoft Windows. At the time of writing 5.72% of all Microsoft Windows installations are still running Windows XP, despite mainstream support ending in 2009 and being superseded by 4 software generations, Vista, 7, 8 and 10.

Thirdly, the signaling methodology is widely misinterpreted to mean the hash power is voting on a proposal and it seems difficult to correct this misunderstanding in the wider community. The hash powers' role is to select valid transactions, and to extend the blockchain with valid blocks. Fully validating economic nodes ensure that blocks are valid. Nodes therefore define validity according to the software they run, but miners decide what already valid transactions gets included in the block chain.

As such, soft forks rules are actually always enforced by the nodes, not the miners. Miners of course can opt-out by simply not including transactions that use the new soft fork feature, but they cannot produce blocks that are invalid to the soft fork. The P2SH soft fork is a good example of this, where non-upgraded miners would see P2SH as spendable without a signature and consider them valid. If such an transaction were to be included in a block, the block would be invalid and the miner would lose the block reward and fees.

So-called "censorship" soft forks do not require nodes to opt in, because >51% of the hash power already have the ability to orphan blocks that contain transactions they have blacklisted. Since this is not a change in validity, nodes will accept the censored block chain automatically.

The fourth problem with supermajority hash power signaling is it draws unnecessary attention to miners which can become unnecessarily political. Already misunderstood as a vote, miners may feel pressure to "make a decision" on behalf of the community: who is and isn't signalling becomes a huge public focus and may put pressures onto miners they are unprepared for. Some miners may not be in a position to upgrade, or may prefer not to participate in the soft fork which is their right. However, that miner may now become a lone reason that vetoes activation for everyone, where the soft fork is an opt-in feature! This situation seems to be against the voluntary nature of the Bitcoin system where participation at all levels is voluntary and kept honest by well balanced incentives.

Since miners already have the protocol level right to select whatever transaction they prefer (and not mine those they don't), it would be better if a miner could chose to not participate in triggering activation of something they won't use, but, without being a veto to the process (and all the ire they may have to experience as a consequence).

The alternative discussed here is "flag day activation" where nodes begin enforcement at a predetermined time in the future. This method needs a longer lead time than a hash power based activation trigger, but offers a number of advantages and perhaps provides a better tradeoff.

Soft forks are still entirely optional to use post activation. For example, with P2SH, many participants in the Bitcoin ecosystem still do not use P2SH. Only 11% of bitcoins[2] are stored in P2SH addresses at the time of writing. Miners are free to not mine P2SH transactions, however, the incentives are such that miners should still validate transactions so they don't accidentally include invalid transactions and cause their block to be rejected. As an additional safety measure for well designed soft forks, relay policy rules prevent non-standard and invalid transactions from being relayed and mined by default; a miner would have to purposefully mine an invalid transaction, which is against their own economic interest.

Since the incentives of the Bitcoin system rely on self validation, economic nodes (miners and users) should always remain safe by ensuring their nodes either validate the current rules, or, they can place their network behind a full node that will filter out invalid transactions and blocks at the edge of their network (so called firewall or border nodes).

A user activated soft fork is permissive. Miners do not have to produce new version blocks and non-upgraded miners' blocks will not be orphaned as was the case with IsSuperMajority soft forks (e.g. BIP34, BIP66, BIP65-CLTV) which made it a compulsory upgrade for miners.

BIP9 "versionbits" soft fork activation method is also permissive in so far as non-upgraded miners are not forced to upgrade after activation because their blocks wont be orphaned. A recent case was the "CSV" soft fork that activated BIP68, BIP112 and BIP113. As such, the CSV soft fork allows non-upgraded miners to continue mining so long as they didn't produce invalid blocks.

Miners always retain discretion on which transactions to mine. However, regardless of whether they actively include transactions using the new soft fork feature, or not, the incentive for hash power to upgrade in order to validate is strong: if they do not, they could be vulnerable to a rogue miner willing to waste 12.5BTC to create an invalid block, which may cause non-validating miners to build on an invalid chain similar to the BIP66 incident. Validation has always had a strong requirement.

A user activated soft fork is win-win because it adds an option that some people want that does not detract from other peoples' enjoyment. Even if only 10% of users ever wanted a feature, so long as the benefit outweighed the technical risks, it would not be rational to deny others the ability to opt-in.

My suggestion is to have the best of both worlds. Since a user activated soft fork needs a relatively long lead time before activation, we can combine with BIP9 to give the option of a faster hash power coordinated activation or activation by flag day, whichever is the sooner. In both cases, we can leverage the warning systems in BIP9. The change is relatively simple, adding an activation-time parameter which will transition the BIP9 state to LOCKED_IN before the end of the BIP9 deployment timeout.

You can find the proposal here https://gist.github.com/shaolinfry/0f7d1fd22743bb966da0c0b1682ea2ab

References:

[1]: https://bitcoin.org/en/alert/2015-07-04-spv-mining
[2]: http://p2sh.info/dashboard/db/p2sh-statistics?from=1472043312917&to=1488030912918

_______________________________________________
bitcoin-dev mailing list
bitcoin-dev@lists.linuxfoundation.org
https://lists.linuxfoundation.org/mailman/listinfo/bitcoin-dev