Blocksize and off-chain transactions

> The very statement that we're willing to increase the blocksize as our
> solution to increased transaction volume rather go down the path of
> off-chain transactions is incredibly controversial.

I really don't understand this either/or mentality.

OF COURSE we're going to raise the block size limit. Limiting the main blockchain to single-digit transactions-per-second is not an option, the vision FOREVER has been to scale it up.

And OF COURSE there will be off-chain transactions-- at the very least, we need them for "instantly confirmed" transactions.

But lets table that whole discussion until 0.8.1 is out the door.

On Wed, Mar 13, 2013 at 01:01:43PM -0400, Gavin Andresen wrote:

> > The very statement that we're willing to increase the blocksize as our
> > solution to increased transaction volume rather go down the path of
> > off-chain transactions is incredibly controversial.
>
> I really don't understand this either/or mentality.

You said it best yourself:

10:48 < gavinandresen> Luke-Jr: argument for another day, but I can almost guarantee that the blocksize limit will be raised in less than 2 years, just based on pressure from the big businesses using the chain (and no, NOT satoshidice)

Decentralization offers big businesses nothing; they're a regulation target already by virtue of size alone.

> OF COURSE we're going to raise the block size limit. Limiting the main
> blockchain to single-digit transactions-per-second is not an option,
> the vision FOREVER has been to scale it up.
>
> And OF COURSE there will be off-chain transactions-- at the very
> least, we need them for "instantly confirmed" transactions.
>
> But lets table that whole discussion until 0.8.1 is out the door.
>
> --
> --
> Gavin Andresen
>

Please note that it was not 0.8 that had issues, but 0.7(and downwards).

I really think changing features in 0.8 aiming for a fluffy limit to avoid lock object errors on 0.7 is the wrong way to go, and it will never cover for a similar situations in the future.

Instead I would like to propose a setup for "considerate mining": * Run pools either on newest or second newest version (up to you depending on which features you like as a pool admin) - say e.g. 0.8 * Connect to the rest of the bitcoin network _only_ through a node of the other version - say e.g. 0.7

This guarantees that no blocks will get into the network that will not be accepted by both 0.8 and 0.7. Those two versions together should add up to say >90%.

Once everyone else (90%) have upgraded to the newest, (0.8), drop the 0.7 and start to introduce 0.9 instead.

/M

On Wednesday, March 13, 2013 6:01:02 PM Michael Gronager wrote:

> Please note that it was not 0.8 that had issues, but 0.7(and downwards).

While 0.7 and earlier do have issues, they also define the Bitcoin protocol. 0.8's failure to comply with the protocol is an issue.

On Wed, Mar 13, 2013 at 07:01:02PM +0100, Michael Gronager wrote:

> Please note that it was not 0.8 that had issues, but 0.7(and downwards).
>
> I really think changing features in 0.8 aiming for a fluffy limit to avoid lock object errors on 0.7 is the wrong way to go, and it will never cover for a similar situations in the future.

The current behaviour in 0.7 and before is indeed broken, and we cannot afford to keep that as an implicitly-defined rule on the network. I fully agree with that, but it will require a hardfork.

But we cannot just drop support for old nodes. It is completely unreasonable to put the _majority_ of the network on a fork, without even as much as a discussion about it. "Oh, you didn't get the memo? The rules implemented in your client are outdated." - that is not how Bitcoin works: the network defines the rules.

IMHO, the way to go is first get a 0.8.1 out that mimics the old behaviour - just as a stopgap solution. That will allow miners to safely use 0.8-based code at least. We can also get patches for 0.7 and previous branches that fix the lock limit issue, but enforce the same limit as 0.8.1 does, simply as band-aid for those who do not yet wish to upgrade to 0.8.

Finally, we'll have to schedule a hard fork to drop the 0.8.1 limit. This is something that requires widespread community consensus - far more than just miners and developers - but as this is about fixing a bug that would otherwise prevent most evolution, I hope it can be a very non-controversial one. To that end, I would prefer to limit this hard fork to only direct bug fixes, and leave the block limit issue for later. By now, I think it is clear that a hard fork will be inevitable, and by doing one, I think we can learn a lot that can make later ones easier.

On Wed, Mar 13, 2013 at 07:28:06PM +0100, Pieter Wuille wrote:

> IMHO, the way to go is first get a 0.8.1 out that mimics the old
> behaviour - just as a stopgap solution.

Presumably not emulate too precisely, at least if your initial report that the block caused 0.7 to 'get stuck' was correct. A network that has a mix of 0.8.1 nodes (which will reject the block) and 0.7 nodes (which will hang when receiving the block?) would appear to remove the fork risk. Is it obvious that no other serious problems remain in such a network?

(Although I note your proposal to patch 0.7 too, so hopefully the network wouldn't remain in that state for very long)

roy

On Wed, Mar 13, 2013 at 2:28 PM, Pieter Wuille <pieter.wuille@gmail•com>wrote:

> But we cannot just drop support for old nodes. It is completely
> unreasonable to put the
> _majority_ of the network on a fork, without even as much as a discussion
> about it.
> "Oh, you didn't get the memo? The rules implemented in your client are
> outdated." - that
> is not how Bitcoin works: the network defines the rules.
> ...
> Finally, we'll have to schedule a hard fork to drop the 0.8.1 limit. This
> is something
> that requires widespread community consensus - far more than just miners
> and developers

The way I've started thinking about it is that there is a market for securing a payment network. In that market you have consumers (users of bitcoin) and providers (miners). It's not clear to me that if the overwhelming majority of miners stayed on 0.8 that the 0.7 fork wouldn't have still won out in the long run because effectively what you would have had is a situation where the providers abandon a large portion of their customers (0.7 users) and start providing a service that is in much less demand. Would everyone have upgraded to 0.8? Maybe, but maybe not. Maybe many people would have made the rational decision to stay on earlier versions and the small minority of miners that choose to service the 0.7 fork could have earned more Bitcoin on that fork...and maybe in the long run, the majority of miners on 0.8 would realize this situation and start to trickle back over to the 0.7 fork. The flip side of the equation is that the users have a pretty compelling reasons to use the services of the most secure network (less risk of double spends). So, the users very well could have made the rational decision to consume the services of the most powerful network and made the switch to 0.8.

What happened in reality is that the majority of the mining community made the rational decision to service the largest pool of customers (0.8 as well as 0.7 and earlier users). It was rational because the risk of servicing only the 0.8 users would have been much greater.

Because of this dynamic, I doubt there would ever be multiple forks of any consequence in permanent coexistence. I would even go so far as to say that at this point, a successful competitor to Bitcoin would have to arise as a fork of the UTXOs in the block chain (even if the competitor didn't even use a block chain). That competitor might even have to begin life in lock step co-existence with Bitcoin, recognizing all Bitcoin transactions for some period of time while it attempts to gain market share.

I hear consensus that at some point we need a hardfork (== creating blocks that will not be accepted by <0.7 clients).

Miners generate block, hence they are the ones who should filter themselves though some consensus.

> But we cannot just drop support for old nodes. It is completely unreasonable to put the
> _majority_ of the network on a fork, without even as much as a discussion about it.
> "Oh, you didn't get the memo? The rules implemented in your client are outdated." - that
> is not how Bitcoin works: the network defines the rules.

Consensus was rapidly reached a day ago: To ensure the majority (all of?) the network could accept the blocks mined, and not just 0.8. This was the right decision! Too many was dependent on <=0.7

So, the question is not if, but when to do a hardfork. We need to define and monitor the % of nodes running different versions (preferably a weighted average - some nodes, like e.g. blockchain.info & mtgox serve many...). Once there was the rowit bitcoinstatus page - do we have another resource for this ?

Then the second question is how to ensure we don't create a fork again? Pieter (and others?) are of the opinion that we should mimic a 0.7 lock-object-starvation-reject-rule. I don't like this for three reasons: 1. I find it hard to ensure we have actually coined the bug precisely 2. I expect that similar issues will happen again 3. The current issue was between two versions, but in the future it could be between two implementations - then trying implement or even to coordinate strange rules becomes very unlikely.

Hence the scheme for "considerate mining" - it is the only scheme that guarantees 100% that no block are released that will not be accepted by a supermajority of the install base.

Another nice thing about it - it requires no development :)

So simply run in serial in front of all considerate miners nodes of different versions until a certain threshold of the install base is reached.

/M