[bitcoin-dev] Proposal: Soft Fork Threshold Signaling

[bitcoin-dev] Proposal: Soft Fork Threshold Signaling

[Thomas Voegtlin]

Disclaimer: I am fully supportive of Segregated Witness and its
activation by users through BIP148. However, I also believe that a
soft fork would be less risky if it was initially activated by miners,
before the date set in BIP148. This proposal is not intended to
replace UASF, but to mitigate the risks.

The following idea might already have been proposed and discussed
elsewhere. If that is the case, I am sorry for the noise.


Background
==========

BIP9 requires 95% of miner hashrate support in order to activate a
soft-fork. So far, the lack of miner consensus about Segwit has been
frustrating both users and developers. This had led some users to
propose a soft fork activation regardless of the expressed level of
miner support (UASF, BIP148).

There are many risks associated with UASF. If the fork is activated
with less than 50% of the hashing power, the blockchain will have two
competing branches. In addition, if the hashrate on the forking branch
is very low, that branch will be exposed to attacks, where non-empty
blocks are systematically orphaned by adverse miners. This threat may
be a strong deterrent for miners willing to support the fork.

The main argument in favor of UASF is that users, not miners, give its
value to Bitcoin. Therefore, users and markets should have the power
to decide which branch of the fork has the most value, and
profit-driven miners should follow. If the soft-forking branch is
valued more than the non-forking branch, it will end up attracting a
majority of the hashing power, and the non-forking branch will
eventually be orphaned.

Feedback through markets, however, will only work if the forking
branch can effectively be used. If the forking branch is rendered
unusable by adverse miners, there is little chance the new coins will
ever reach markets. To make things worse, profit-driven miners might
adopt a passive attitude and decide to mine on the forking branch only
once a proper price has been set by markets, or only once they see
that it has enough hashing power to be usable. Thus, the lack of
hashrate information prior to the soft fork increases the risk.

On the other hand, if a soft fork was initiated with more than 33% of
the hashing power, then it would probably be viable, because the
remaining two thirds of the hashing power cannot successfully be
allocated to mine blocks on the non-forking branch and to orphan
blocks on the forking branch. Therefore, users will be able to move
coins on the forking branch, and markets will be able to set a price
on these coins, thus creating the feedback needed by profit-driven
miners.

Today about 30% of the hashing power are signaling their intention to
activate Segwit using BIP9. This hashrate is very close to the 33%
threshold, and it would probably be enough to initiate a viable soft
fork; indeed we can expect additional hashing power to be gained from
miners mining on both branches of the fork.

However, nothing suggests that a soft fork triggered with 30% of the
hashrate would be followed by the miners who are currently signaling
Segwit using BIP9. BIP9 signaling means that these miners are willing
to soft fork if support reaches 95%; it does not say anything about
their intentions if support is as low as 30%. In other words, BIP9
signaling does allow miners to properly signal their intentions.


BIP9 signaling
==============

The activation threshold is part of the semantics of BIP9. Miners who
use BIP9 do not only signal their support for a soft fork; they also
signal to other miners that they will activate the soft fork if and
only if support reaches 95%.

Some of these miners might actually be willing to activate a soft fork
with a lower support, even at the cost of creating two chains. Other
miners might not be supportive of that idea, because they consider
that the danger of their blocks being orphaned is too high.

The problem is that this information, at which level of support miners
are willing to initiate a soft fork, is not available. Thus, miners
who are willing to initiate a soft fork at a lower hashrate cannot
coordinate their action.


Proposal: Soft Fork Threshold Signaling
=======================================

Miners signal the threshold at which they are willing to activate a
soft fork. The value of the threshold is published in the coinbase
transaction of the block, with the corresponding version bit.

Miners activate a soft fork if their threshold has been reached over
the last retargeting period. For example, if 504 of 2016 blocks signal
a soft fork with a threshold equal or lower to 25%, then the soft fork
is activated by these miners.

If no activation threshold is reached, the current BIP9 signaling rate
indicator is replaced by a distribution of signaling rates per
threshold. The public availability of threshold information allows
miners to adjust their own threshold, and to optimize their chances of
activating the soft fork.


UASF
====

Even if the soft fork is not activated by miners, this proposal will
reduce the risks associated to a user activated soft fork (UASF). The
public availability of hashrate threshold information prior to the
soft fork will help miners decide whether they should join the fork
right after it has been activated, before price information is
available.


Vulnerabilities
===============

This proposal has similar vulnerabilities as BIP9: it is susceptible
to fake signaling by miners, and to miners withholding hashing power
before the fork.

Re: [bitcoin-dev] Proposal: Soft Fork Threshold Signaling

[Sancho Panza]

[-- Attachment #1: Type: text/plain, Size: 560 bytes --]

Thomas,

I wonder if you've seen my proposal on how to make BIP9 more configurable:
https://github.com/sanch0panza/bips/blob/bip-genvbvoting/bip-genvbvoting.mediawiki

This could be extended with a coinbase signaling feature as you suggest.
This could include parameter information for forks which a miner is signaling, for coordination.

Currently I've not included something like this, but it might make a nice addition.
One problem is the limited space in coinbase for embedding data on the large number of possible independent deployments.

Regards,
Sancho

[-- Attachment #2: Type: text/html, Size: 840 bytes --]

Re: [bitcoin-dev] Proposal: Soft Fork Threshold Signaling

[Thomas Voegtlin]

Hi Sancho,

I saw your proposal. However, my point is that the threshold should be
part of the signaling, and not fixed in the soft-fork proposal.

I agree that coinbase space might be a limitation.

To avoid this, I realize that the threshold could be encoded in the
version bits. We have 32 version bits, and the top 3 bits must be set to
001 in BIP9. In order to extend BIP9 in a backward compatible way, we
could set these 3 top bits to 010, and use the 29 remaining bits for
soft fork signaling.

If we use 7 bits per soft-fork proposal, we have enough space to encode
four simultaneous soft-fork proposals, and sub-percent granularity for
the threshold (2^7=128).



Le 13/04/2017 à 16:17, Sancho Panza a écrit :
> Thomas,
> 
> I wonder if you've seen my proposal on how to make BIP9 more configurable:
> https://github.com/sanch0panza/bips/blob/bip-genvbvoting/bip-genvbvoting.mediawiki
> 
> This could be extended with a coinbase signaling feature as you suggest.
> This could include parameter information for forks which a miner is signaling, for coordination.
> 
> Currently I've not included something like this, but it might make a nice addition.
> One problem is the limited space in coinbase for embedding data on the large number of possible independent deployments.
> 
> Regards,
> Sancho
> 

Re: [bitcoin-dev] Proposal: Soft Fork Threshold Signaling

[Sancho Panza]

[-- Attachment #1: Type: text/plain, Size: 1112 bytes --]

> However, my point is that the threshold should be [...] not fixed in the soft-fork proposal

My proposal makes it configurable (as well as window size, grace period etc.)

> I agree that coinbase space might be a limitation.

I still like the coinbase idea though - more than using up the BIP9 versionbits range for verbose signaling.

BIP9 (and other proposals which use those 29 versionbits) currently assume that the participants on the network will coordinate in some form or other, to agree on what the bits mean (in terms of change deployments).

It would be very easy to also agree on a set of "standard" threshold levels and map those onto e.g. 1 byte.

Then, in the coinbase, one could have pairs of bit numbers and bytes, e.g. "/1A/2B/3C/" where the bytes values corresponding to 'A', 'B', 'C', ... are standardized deployment schedules that people find useful.
So a BIP9 conformant schedule could be A = 95% / 2016 window,
while B = 75%/2016, etc.

This would be quite a compact yet still readable signaling. The space of values is large enough that I doubt we'd see much contention.

Regards
Sancho

[-- Attachment #2: Type: text/html, Size: 1408 bytes --]

Re: [bitcoin-dev] Proposal: Soft Fork Threshold Signaling

[Thomas Voegtlin]

I think it is better not to use the coinbase, because it might collide
with other proposals, and because coinbase is not part of the block header.

I agree that a small set of standard threshold may be sufficient; a one
percent resolution is probably not needed. If we use 4 bits we can
encode 15 different thresholds + zero (meaning no support). For example
we can have all thresholds between 25% and 95% separated by 5%.

Using 4 bits per soft-fork proposal leaves enough room to fit 7
simultaneous proposals in version bits. That should be plenty.

> 
> I still like the coinbase idea though - more than using up the BIP9 versionbits range for verbose signaling.
> 
> BIP9 (and other proposals which use those 29 versionbits) currently assume that the participants on the network will coordinate in some form or other, to agree on what the bits mean (in terms of change deployments).
> 
> It would be very easy to also agree on a set of "standard" threshold levels and map those onto e.g. 1 byte.
> 
> Then, in the coinbase, one could have pairs of bit numbers and bytes, e.g. "/1A/2B/3C/" where the bytes values corresponding to 'A', 'B', 'C', ... are standardized deployment schedules that people find useful.
> So a BIP9 conformant schedule could be A = 95% / 2016 window,
> while B = 75%/2016, etc.
> 
> This would be quite a compact yet still readable signaling. The space of values is large enough that I doubt we'd see much contention.
> 
> Regards
> Sancho
>