[bitcoin-dev] Hypothetical 2 MB hardfork to follow BIP148

Discussion:

Jorge Timón via bitcoin-dev

2017-05-30 13:27:58 UTC

Why not simply remove the (redundant after sw activation) 1 mb size
limit check and increasing the weight limit without changing the
discount or having 2 limits?

On Wed, May 24, 2017 at 1:07 AM, Erik Aronesty via bitcoin-dev

Personally, I would prefer if a 2MB lock-in that uses BIP103 for the timing.
I think up to 20% per year can be absorbed by averages in bandwidth/CPU/RAM
growth, of which bandwidth seems the most constraining.
- Erik
On Tue, May 23, 2017 at 4:23 PM, Luke Dashjr via bitcoin-dev

In light of some recent discussions, I wrote up this BIP for a real 2 MB
block
size hardfork following Segwit BIP148 activation. This is not part of any
agreement I am party to, nor anything of that sort. Just something to
throw
out there as a possible (and realistic) option.
Note that I cannot recommend this to be adopted, since frankly 1 MB blocks
really are still too large, and this blunt-style hardfork quite risky even
with consensus. But if the community wishes to adopt (by unanimous
consensus)
a 2 MB block size hardfork, this is probably the best way to do it right
now.
The only possible way to improve on this IMO would be to integrate it into
MMHF/"spoonnet" style hardfork (and/or add other unrelated-to-block-size
HF
improvements).
I have left Author blank, as I do not intend to personally champion this.
Before it may be assigned a BIP number, someone else will need to step up
to
take on that role. Motivation and Rationale are blank because I do not
personally think there is any legitimate rationale for such a hardfork at
this
time; if someone adopts this BIP, they should complete these sections. (I
can
push a git branch with the BIP text if someone wants to fork it.)
<pre>
BIP: ?
Layer: Consensus (hard fork)
Title: Post-segwit 2 MB block size hardfork
Author: FIXME
Comments-Summary: No comments yet.
Comments-URI: ?
Status: Draft
Type: Standards Track
Created: 2017-05-22
License: BSD-2-Clause
</pre>
==Abstract==
Legacy Bitcoin transactions are given the witness discount, and a block
size
limit of 2 MB is imposed.
==Copyright==
This BIP is licensed under the BSD 2-clause license.
==Specification==
Upon activation, a block size limit of 2000000 bytes is enforced.
The block weight limit remains at 4000000 WU.
all witness data, including both scriptSig (used by pre-segwit inputs) and
segwit witness data, is measured as 1 weight-unit (WU), while all other
data
in the block is measured as 4 WU.
The witness commitment in the generation transaction is no longer
required,
and instead the txid merkle root in the block header is replaced with a
hash
1. The witness reserved value.
2. The witness merkle root hash.
3. The transaction ID merkle root hash.
The maximum size of a transaction stripped of witness data is limited to 1
MB.
===Deployment===
This BIP is deployed by flag day, in the block where the median-past time
surpasses 1543503872 (2018 Nov 29 at 15:04:32 UTC).
It is assumed that when this flag day has been reached, Segwit has been
activated via BIP141 and/or BIP148.
==Motivation==
FIXME
==Rationale==
FIXME
==Backwards compatibility==
This is a hardfork, and as such not backward compatible.
It should not be deployed without consent of the entire Bitcoin community.
Activation is scheduled for 18 months from the creation date of this BIP,
intended to give 6 months to establish consensus, and 12 months for
deployment.
==Reference implementation==
FIXME
_______________________________________________
bitcoin-dev mailing list
https://lists.linuxfoundation.org/mailman/listinfo/bitcoin-dev

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

Jacob Eliosoff via bitcoin-dev

2017-05-30 20:10:04 UTC

Permalink

I'd like to know this too. Is it just that a 4MB blockmaxweight would
theoretically allow ~4MB blocks (if ~all witness data), which is too big?
Or is there a more subtle reason we still need blockmaxsize after a HF?

On May 30, 2017 9:28 AM, "Jorge TimÃ³n via bitcoin-dev" <
bitcoin-***@lists.linuxfoundation.org> wrote:

Why not simply remove the (redundant after sw activation) 1 mb size
limit check and increasing the weight limit without changing the
discount or having 2 limits?

On Wed, May 24, 2017 at 1:07 AM, Erik Aronesty via bitcoin-dev

Personally, I would prefer if a 2MB lock-in that uses BIP103 for the timing.
I think up to 20% per year can be absorbed by averages in

bandwidth/CPU/RAM

growth, of which bandwidth seems the most constraining.
- Erik
On Tue, May 23, 2017 at 4:23 PM, Luke Dashjr via bitcoin-dev

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

Mark Friedenbach via bitcoin-dev

2017-05-30 21:24:20 UTC

Permalink

The 1MB classic block size is not redundant after segwit activation.
Segwit prevents the quadratic hashing problems, but only for segwit
outputs. The 1MB classic block size prevents quadratic hashing
problems from being any worse than they are today.

Mark

On Tue, May 30, 2017 at 6:27 AM, Jorge Timón via bitcoin-dev

Post by Jorge TimÃ³n via bitcoin-dev
Why not simply remove the (redundant after sw activation) 1 mb size
limit check and increasing the weight limit without changing the
discount or having 2 limits?
On Wed, May 24, 2017 at 1:07 AM, Erik Aronesty via bitcoin-dev

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

Jorge Timón via bitcoin-dev

2017-05-30 22:26:20 UTC

Permalink

My understanding is that you cannot possibly violate the 1 MB block
size rule without also violating the 4 MB weight rule.
Regarding size alone, the only check we care about if we accept segwit is:

https://github.com/bitcoin/bitcoin/blob/master/src/validation.cpp#L2891 [size4]

If that doesn't fail due to excessive non-witness data, then there's no way that

https://github.com/bitcoin/bitcoin/blob/master/src/validation.cpp#L2681 [size1]

would have failed before due to excessive non-witness data.
If I understood it correctly when I was explained, if I remember
correctly, that last check is really just an optimization or a
protection against DoS invalid blocks. If the size without any witness
data is bigger than 1/4 the max_weight, then the max_weight check is
certain to fail as well without having to look at any witness data at
that validation stage (assuming the failure is due to excessive
non-witness data).

I think you are not referring to the 1 mb size limit but to related
one for sigops:

https://github.com/bitcoin/bitcoin/blob/master/src/validation.cpp#L2704
[sigops1]

whose segwit parallel is in:

https://github.com/bitcoin/bitcoin/blob/master/src/validation.cpp#L1661
[sigops4]

I believe the situation is similar in checking before knowing anything
about the witness data just in case that's already too much. In fact,
here is clearer because MAX_BLOCK_SIGOPS_COST is used for both (and
WITNESS_SCALE_FACTOR is used for the optimization case).

So what I would do in a hardfork after segwit activation would be to
simply equal MAX_BLOCK_BASE_SIZE=MAX_BLOCK_WEIGHT/WITNESS_SCALE_FACTOR
for size1, and increase MAX_BLOCK_WEIGHT and MAX_BLOCK_ SIGOPS_COST
proportionally for size4 and sigops4 respectively (well, the sigops
const for sigops1 as well).

If I understood segwit correctly, I believe that even though it is not
activated yet, you could remove both the size1 and sigops1 checks and
your node would still not accept invalid blocks by pre-bip141 rules,
your node would just spend more time on invalid blocks due to
currently excessive size/sigops, because it would only realize at a
later validation stage. Sorry for the redundancy about the validation
stage.

But it is not unlikely that I'm missing something. If I am wrong about
this I am spreading misinformation about segwit in several channels,
so I'm very interested in corrections to my statements in this mail.

Post by Mark Friedenbach via bitcoin-dev
The 1MB classic block size is not redundant after segwit activation.
Segwit prevents the quadratic hashing problems, but only for segwit
outputs. The 1MB classic block size prevents quadratic hashing
problems from being any worse than they are today.
Mark
On Tue, May 30, 2017 at 6:27 AM, Jorge Timón via bitcoin-dev

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

James MacWhyte via bitcoin-dev

2017-05-30 23:50:01 UTC

Permalink

Post by Mark Friedenbach via bitcoin-dev
The 1MB classic block size prevents quadratic hashing
problems from being any worse than they are today.

Add a transaction-size limit of, say, 10kb and the quadratic hashing
problem is a non-issue. Donezo.

Jorge Timón via bitcoin-dev

2017-05-31 01:22:44 UTC

Permalink

Post by Mark Friedenbach via bitcoin-dev
The 1MB classic block size prevents quadratic hashing
problems from being any worse than they are today.

Add a transaction-size limit of, say, 10kb and the quadratic hashing problem
is a non-issue. Donezo.

Why is it https://github.com/bitcoin/bitcoin/blob/master/src/validation.cpp#L1661
not enough at this point?
Why the need for a transaction size limit?

Luke Dashjr via bitcoin-dev

2017-05-31 04:14:25 UTC

Permalink

Post by Jorge TimÃ³n via bitcoin-dev
Why is it
https://github.com/bitcoin/bitcoin/blob/master/src/validation.cpp#L1661
not enough at this point?
Why the need for a transaction size limit?

Because the bottleneck is hashing the transaction, which costs (in CPU time)
based on its size. Maybe it would make sense to factor sigops into the limit,
though?

Post by Jorge TimÃ³n via bitcoin-dev

Post by James MacWhyte via bitcoin-dev
Add a transaction-size limit of, say, 10kb and the quadratic hashing
problem is a non-issue. Donezo.

That would invalidate any pre-signed transactions that are currently out
there. You can't just change the rules out from under people.

Make it 100kB and I think we'd be okay. Those have always been policy-
forbidden so there should be no expectation they'd be acceptable in the
future.

While we're at it, I suggest also specifying a minimum transaction size as
well. The raw minimum possible is 60 bytes, but any sane output would need at
least a hash, so I'd say make the minimum be 8 (60 + 160-bit hash) bytes?

Luke

Jean-Paul Kogelman via bitcoin-dev

2017-05-31 01:09:26 UTC

Permalink

That would invalidate any pre-signed transactions that are currently out there. You can't just change the rules out from under people.

Post by Mark Friedenbach via bitcoin-dev
The 1MB classic block size prevents quadratic hashing
problems from being any worse than they are today.

Jacob Eliosoff via bitcoin-dev

2017-05-31 03:07:43 UTC

Permalink

Maybe there's some hole in Jorge's logic and scrapping blockmaxsize has
quadratic hashing risks, and maybe James' 10KB is too ambitious; but even
if so, a simple 1MB tx size limit would clearly do the trick. The broader
point is that quadratic hashing is not a compelling reason to keep
blockmaxsize post-HF: does someone have a better one?

On May 30, 2017 9:46 PM, "Jean-Paul Kogelman via bitcoin-dev" <

Post by Jean-Paul Kogelman via bitcoin-dev
That would invalidate any pre-signed transactions that are currently out
there. You can't just change the rules out from under people.
On May 30, 2017, at 4:50 PM, James MacWhyte via bitcoin-dev <

Post by Mark Friedenbach via bitcoin-dev
The 1MB classic block size prevents quadratic hashing
problems from being any worse than they are today.

Add a transaction-size limit of, say, 10kb and the quadratic hashing
problem is a non-issue. Donezo.
_______________________________________________
bitcoin-dev mailing list
https://lists.linuxfoundation.org/mailman/listinfo/bitcoin-dev
_______________________________________________
bitcoin-dev mailing list
https://lists.linuxfoundation.org/mailman/listinfo/bitcoin-dev

Jared Lee Richardson via bitcoin-dev

2017-06-02 19:40:36 UTC

Permalink

Maybe there's some hole in Jorge's logic and scrapping blockmaxsize has quadratic hashing risks, and maybe James' 10KB is too ambitious; but even if so, a simple 1MB tx size limit would clearly do the trick. The broader point is that quadratic hashing is not a compelling reason to keep blockmaxsize post-HF: does someone have a better one?

I think this is exactly the right direction to head. There are
arguments to be made for various maximum sizes... Maybe the limit
could be set to 1mb initially, and at a distant future block
height(years?) automatically drop to 500kb or 100kb? That would give
anyone with existing systems or pre-signed transactions several years
to adjust to the change. Notification could (?possibly?) be done with
a non-default parameter that must be changed to continue to use 100kb
- <1mb transactions, so no one running modern software could claim
they were not informed when that future date hits.

I don't see any real advantages to continuing to support transactions
larger than 100kb excepting the need to update legacy use cases /
already signed transactions.

On Tue, May 30, 2017 at 8:07 PM, Jacob Eliosoff via bitcoin-dev

Maybe there's some hole in Jorge's logic and scrapping blockmaxsize has
quadratic hashing risks, and maybe James' 10KB is too ambitious; but even if
so, a simple 1MB tx size limit would clearly do the trick. The broader
point is that quadratic hashing is not a compelling reason to keep
blockmaxsize post-HF: does someone have a better one?
On May 30, 2017 9:46 PM, "Jean-Paul Kogelman via bitcoin-dev"

Post by Mark Friedenbach via bitcoin-dev
The 1MB classic block size prevents quadratic hashing
problems from being any worse than they are today.

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

Nathan Cook via bitcoin-dev

2017-06-12 16:27:14 UTC

Permalink

The problem with >100kb transactions isn't that there are a lot of
already-signed transactions out there, or even that there are good use
cases for them, but that such transactions and use cases could exist, and
there is no way of disallowing them without possibly costing someone a lot
of money. Slowly reducing the limit doesn't really solve this problem.

I propose to make them hard enough to confirm that no-one will use them
without a very good reason. Just require that any block containing an
outsized transaction be mined at a greater difficulty â quadratically
greater. Such a block is more expensive *for the block's miner*, not just
for the other miners, or for other nodes. Anyone who really, really needs
to use a 400kb transaction can pay a miner to mine it.

Quadratic hashing isn't risky when it is inherently limited by a
corresponding reduction in the rate at which the "bad" blocks can be
generated. That said, there's nothing I can see which is actually good
about large, expensive to validate transactions, and so >1MB transactions
should remain invalid, as they are today.

Nathan Cook

On 2 June 2017 at 22:40, Jared Lee Richardson via bitcoin-dev <

Post by Jacob Eliosoff via bitcoin-dev

Post by Jacob Eliosoff via bitcoin-dev
Maybe there's some hole in Jorge's logic and scrapping blockmaxsize has

quadratic hashing risks, and maybe James' 10KB is too ambitious; but even
if so, a simple 1MB tx size limit would clearly do the trick. The broader
point is that quadratic hashing is not a compelling reason to keep
blockmaxsize post-HF: does someone have a better one?
I think this is exactly the right direction to head. There are
arguments to be made for various maximum sizes... Maybe the limit
could be set to 1mb initially, and at a distant future block
height(years?) automatically drop to 500kb or 100kb? That would give
anyone with existing systems or pre-signed transactions several years
to adjust to the change. Notification could (?possibly?) be done with
a non-default parameter that must be changed to continue to use 100kb
- <1mb transactions, so no one running modern software could claim
they were not informed when that future date hits.
I don't see any real advantages to continuing to support transactions
larger than 100kb excepting the need to update legacy use cases /
already signed transactions.
On Tue, May 30, 2017 at 8:07 PM, Jacob Eliosoff via bitcoin-dev

Post by Jacob Eliosoff via bitcoin-dev
Maybe there's some hole in Jorge's logic and scrapping blockmaxsize has
quadratic hashing risks, and maybe James' 10KB is too ambitious; but

even if

Post by Jacob Eliosoff via bitcoin-dev
so, a simple 1MB tx size limit would clearly do the trick. The broader
point is that quadratic hashing is not a compelling reason to keep
blockmaxsize post-HF: does someone have a better one?
On May 30, 2017 9:46 PM, "Jean-Paul Kogelman via bitcoin-dev"

Post by Mark Friedenbach via bitcoin-dev
The 1MB classic block size prevents quadratic hashing
problems from being any worse than they are today.

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