Gavin Andresen [ARCHIVE] on Nostr: š Original date posted:2016-02-06 š Original message:Responding to "28 days is ...
š
Original date posted:2016-02-06
š Original message:Responding to "28 days is not long enough" :
I keep seeing this claim made with no evidence to back it up. As I said, I
surveyed several of the biggest infrastructure providers and the btcd lead
developer and they all agree "28 days is plenty of time."
For individuals... why would it take somebody longer than 28 days to either
download and restart their bitcoind, or to patch and then re-run (the patch
can be a one-line change MAX_BLOCK_SIZE from 1000000 to 2000000)?
For the Bitcoin Core project: I'm well aware of how long it takes to roll
out new binaries, and 28 days is plenty of time.
I suspect there ARE a significant percentage of un-maintained full nodes--
probably 30 to 40%. Losing those nodes will not be a problem, for three
reasons:
1) The network could shrink by 60% and it would still have plenty of open
connection slots
2) People are committing to spinning up thousands of supports-2mb-nodes
during the grace period.
3) We could wait a year and pick up maybe 10 or 20% more.
I strongly disagree with the statement that there is no cost to a longer
grace period. There is broad agreement that a capacity increase is needed
NOW.
To bring it back to bitcoin-dev territory: are there any TECHNICAL
arguments why an upgrade would take a business or individual longer than 28
days?
Responding to Luke's message:
On Sat, Feb 6, 2016 at 1:12 AM, Luke Dashjr via bitcoin-dev
> <bitcoin-dev at lists.linuxfoundation.org> wrote:
> > On Friday, February 05, 2016 8:51:08 PM Gavin Andresen via bitcoin-dev
> wrote:
> >> Blog post on a couple of the constants chosen:
> >> http://gavinandresen.ninja/seventyfive-twentyeight
> >
> > Can you put this in the BIP's Rationale section (which appears to be
> mis-named
> > "Discussion" in the current draft)?
>
I'll rename the section and expand it a little. I think standards documents
like BIPs should be concise, though (written for implementors), so I'm not
going to recreate the entire blog post there.
> >
> >> Signature operations in un-executed branches of a Script are not counted
> >> OP_CHECKMULTISIG evaluations are counted accurately; if the signature
> for a
> >> 1-of-20 OP_CHECKMULTISIG is satisified by the public key nearest the top
> >> of the execution stack, it is counted as one signature operation. If it
> is
> >> satisfied by the public key nearest the bottom of the execution stack,
> it
> >> is counted as twenty signature operations. Signature operations
> involving
> >> invalidly encoded signatures or public keys are not counted towards the
> >> limit
> >
> > These seem like they will break static analysis entirely. That was a
> noted
> > reason for creating BIP 16 to replace BIP 12. Is it no longer a concern?
> Would
> > it make sense to require scripts to commit to the total accurate-sigop
> count
> > to fix this?
>
After implementing static counting and accurate counting... I was wrong.
Accurate/dynamic counting/limiting is quick and simple and can be
completely safe (the counting code can be told the limit and can
"early-out" validation).
I think making scripts commit to a total accurate sigop count is a bad
idea-- it would make multisignature signing more complicated for zero
benefit. E.g. if you're circulating a partially signed transaction to that
must be signed by 2 of 5 people, you can end up with a transaction that
requires 2, 3, 4, or 5 signature operations to validate (depending on which
public keys are used to do the signing). The first signer might have no
idea who else would sign and wouldn't know the accurate sigop count.
> >
> >> The amount of data hashed to compute signature hashes is limited to
> >> 1,300,000,000 bytes per block.
> >
> > The rationale for this wasn't in your blog post. I assume it's based on
> the
> > current theoretical max at 1 MB blocks? Even a high-end PC would
> probably take
> > 40-80 seconds just for the hashing, however - maybe a lower limit would
> be
> > best?
>
It is slightly more hashing than was required to validate block number
364,422.
There are a couple of advantages to a very high limit:
1) When the fork is over, special-case code for dealing with old blocks can
be eliminated, because all old blocks satisfy the new limit.
2) More importantly, if the limit is small enough it might get hit by
standard transactions, then block creation code (CreateNewBlock() /
getblocktemplate / or some external transaction-assembling software) will
have to solve an even more complicated bin-packing problem to optimize for
fees paid.
In practice, the 20,000 sigop limit will always be reached before
MAX_BLOCK_SIGHASH.
> >
> >> Miners express their support for this BIP by ...
> >
> > But miners don't get to decide hardforks. How does the economy express
> their
> > support for it? What happens if miners trigger it without consent from
> the
> > economy?
>
"The economy" does support this.
> >
> > If you are intent on using the version bits to trigger the hardfork, I
> suggest
> > rephrasing this such that miners should only enable the bit when they
> have
> > independently confirmed economic support (this means implementations
> need a
> > config option that defaults to off).
>
Happy to add words about economic majority.
Classic will not implement a command-line option (the act of running
Classic is "I opt in"), but happy to add one for a pull request to Core,
assuming Core would not see such a pull request as having any hostile
intent.
>
> >> SPV (simple payment validation) wallets are compatible with this change.
> >
> > Would prefer if this is corrected to "Light clients" or something.
> Actual SPV
> > wallets do not exist at this time, and would not be compatible with a
> > hardfork.
>
Is there an explanation of SPV versus "Light Client" written somewhere more
permanent than a reddit comment or forum post that I can point to?
> >
> >> In the short term, an increase is needed to continue the current
> economic
> >> policies with regards to fees and block space, matching market
> expectations
> >> and preventing market disruption.
> >
> > IMO this sentence is the most controversial part of your draft, and it
> > wouldn't suffer a loss to remove it (or at least make it subjective).
>
Happy to remove.
> > I would also prefer to see any hardfork:
> >
> > 1. Address at least the simple tasks on the hardfork wishlist (eg,
> enable some
> > disabled opcodes; fix P2SH for N-of->15 multisig; etc).
>
Those would be separate BIPs. (according to BIP 1, smaller is better)
After this 2MB bump, I agree we need to agree on a process for the next
hard fork to avoid all of the unnecessary drama.
> 2. Be deployed as a soft-hardfork so as not to leave old nodes entirely
> > insecure.
>
I haven't been paying attention to all of the
"soft-hardfork/hard-softfork/etc" terminology so have no idea what you
mean. Is THAT written up somewhere?
--
--
Gavin Andresen
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linuxfoundation.org/pipermail/bitcoin-dev/attachments/20160206/7c874b34/attachment.html>;
š Original message:Responding to "28 days is not long enough" :
I keep seeing this claim made with no evidence to back it up. As I said, I
surveyed several of the biggest infrastructure providers and the btcd lead
developer and they all agree "28 days is plenty of time."
For individuals... why would it take somebody longer than 28 days to either
download and restart their bitcoind, or to patch and then re-run (the patch
can be a one-line change MAX_BLOCK_SIZE from 1000000 to 2000000)?
For the Bitcoin Core project: I'm well aware of how long it takes to roll
out new binaries, and 28 days is plenty of time.
I suspect there ARE a significant percentage of un-maintained full nodes--
probably 30 to 40%. Losing those nodes will not be a problem, for three
reasons:
1) The network could shrink by 60% and it would still have plenty of open
connection slots
2) People are committing to spinning up thousands of supports-2mb-nodes
during the grace period.
3) We could wait a year and pick up maybe 10 or 20% more.
I strongly disagree with the statement that there is no cost to a longer
grace period. There is broad agreement that a capacity increase is needed
NOW.
To bring it back to bitcoin-dev territory: are there any TECHNICAL
arguments why an upgrade would take a business or individual longer than 28
days?
Responding to Luke's message:
On Sat, Feb 6, 2016 at 1:12 AM, Luke Dashjr via bitcoin-dev
> <bitcoin-dev at lists.linuxfoundation.org> wrote:
> > On Friday, February 05, 2016 8:51:08 PM Gavin Andresen via bitcoin-dev
> wrote:
> >> Blog post on a couple of the constants chosen:
> >> http://gavinandresen.ninja/seventyfive-twentyeight
> >
> > Can you put this in the BIP's Rationale section (which appears to be
> mis-named
> > "Discussion" in the current draft)?
>
I'll rename the section and expand it a little. I think standards documents
like BIPs should be concise, though (written for implementors), so I'm not
going to recreate the entire blog post there.
> >
> >> Signature operations in un-executed branches of a Script are not counted
> >> OP_CHECKMULTISIG evaluations are counted accurately; if the signature
> for a
> >> 1-of-20 OP_CHECKMULTISIG is satisified by the public key nearest the top
> >> of the execution stack, it is counted as one signature operation. If it
> is
> >> satisfied by the public key nearest the bottom of the execution stack,
> it
> >> is counted as twenty signature operations. Signature operations
> involving
> >> invalidly encoded signatures or public keys are not counted towards the
> >> limit
> >
> > These seem like they will break static analysis entirely. That was a
> noted
> > reason for creating BIP 16 to replace BIP 12. Is it no longer a concern?
> Would
> > it make sense to require scripts to commit to the total accurate-sigop
> count
> > to fix this?
>
After implementing static counting and accurate counting... I was wrong.
Accurate/dynamic counting/limiting is quick and simple and can be
completely safe (the counting code can be told the limit and can
"early-out" validation).
I think making scripts commit to a total accurate sigop count is a bad
idea-- it would make multisignature signing more complicated for zero
benefit. E.g. if you're circulating a partially signed transaction to that
must be signed by 2 of 5 people, you can end up with a transaction that
requires 2, 3, 4, or 5 signature operations to validate (depending on which
public keys are used to do the signing). The first signer might have no
idea who else would sign and wouldn't know the accurate sigop count.
> >
> >> The amount of data hashed to compute signature hashes is limited to
> >> 1,300,000,000 bytes per block.
> >
> > The rationale for this wasn't in your blog post. I assume it's based on
> the
> > current theoretical max at 1 MB blocks? Even a high-end PC would
> probably take
> > 40-80 seconds just for the hashing, however - maybe a lower limit would
> be
> > best?
>
It is slightly more hashing than was required to validate block number
364,422.
There are a couple of advantages to a very high limit:
1) When the fork is over, special-case code for dealing with old blocks can
be eliminated, because all old blocks satisfy the new limit.
2) More importantly, if the limit is small enough it might get hit by
standard transactions, then block creation code (CreateNewBlock() /
getblocktemplate / or some external transaction-assembling software) will
have to solve an even more complicated bin-packing problem to optimize for
fees paid.
In practice, the 20,000 sigop limit will always be reached before
MAX_BLOCK_SIGHASH.
> >
> >> Miners express their support for this BIP by ...
> >
> > But miners don't get to decide hardforks. How does the economy express
> their
> > support for it? What happens if miners trigger it without consent from
> the
> > economy?
>
"The economy" does support this.
> >
> > If you are intent on using the version bits to trigger the hardfork, I
> suggest
> > rephrasing this such that miners should only enable the bit when they
> have
> > independently confirmed economic support (this means implementations
> need a
> > config option that defaults to off).
>
Happy to add words about economic majority.
Classic will not implement a command-line option (the act of running
Classic is "I opt in"), but happy to add one for a pull request to Core,
assuming Core would not see such a pull request as having any hostile
intent.
>
> >> SPV (simple payment validation) wallets are compatible with this change.
> >
> > Would prefer if this is corrected to "Light clients" or something.
> Actual SPV
> > wallets do not exist at this time, and would not be compatible with a
> > hardfork.
>
Is there an explanation of SPV versus "Light Client" written somewhere more
permanent than a reddit comment or forum post that I can point to?
> >
> >> In the short term, an increase is needed to continue the current
> economic
> >> policies with regards to fees and block space, matching market
> expectations
> >> and preventing market disruption.
> >
> > IMO this sentence is the most controversial part of your draft, and it
> > wouldn't suffer a loss to remove it (or at least make it subjective).
>
Happy to remove.
> > I would also prefer to see any hardfork:
> >
> > 1. Address at least the simple tasks on the hardfork wishlist (eg,
> enable some
> > disabled opcodes; fix P2SH for N-of->15 multisig; etc).
>
Those would be separate BIPs. (according to BIP 1, smaller is better)
After this 2MB bump, I agree we need to agree on a process for the next
hard fork to avoid all of the unnecessary drama.
> 2. Be deployed as a soft-hardfork so as not to leave old nodes entirely
> > insecure.
>
I haven't been paying attention to all of the
"soft-hardfork/hard-softfork/etc" terminology so have no idea what you
mean. Is THAT written up somewhere?
--
--
Gavin Andresen
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linuxfoundation.org/pipermail/bitcoin-dev/attachments/20160206/7c874b34/attachment.html>;