[bitcoin-dev] Proposed new policy for transactions that depend on other unconfirmed transactions

[Alex Morcos]

I'd like to propose updates to the new policy limits on unconfirmed
transaction chains.

The existing limits in master and scheduled for release in 0.12 are:
Ancestor packages = 100 txs and 900kb total size
Descendant packages = 1000 txs and 2500kb total size

Before 0.12 is released I would like to propose a significant reduction in
these limits. In the course of analyzing algorithms for mempool limiting,
it became clear that large packages of unconfirmed transactions were the
primary vector for mempool clogging or relay fee boosting attacks. Feedback
from the initial proposed limits was that they were too generous anyway.

The proposed new limits are:
Ancestor packages = 25 txs and 100kb total size
Descendant packages = 25 txs and 100kb total size

Based on historical transaction data, the most restrictive of these limits
is the 25 transaction count on descendant packages. Over the period of
April and May of this year (before stress tests), 5.8% of transactions
would have violated this limit alone. Applying all the limits together
would have affected 6.1% of transactions.

Please keep in mind these are policy limits that affect transactions which
depend on other unconfirmed transactions only. They are not a change to
consensus rules and do not affect how many chained txs a valid block may
contain. Furthermore, any transaction that was unable to be relayed due to
these limits need only wait for some of its unconfirmed ancestors to be
included in a block and then it could be successfully broadcast. This is
unlikely to affect the total time from creation to inclusion in a block.
Finally, these limits are command line arguments that can easily be changed
on an individual node basis in Bitcoin Core.

Please give your feedback if you know of legitimate use cases that would be
hindered by these limits.

Thanks,
Alex

On Mon, Sep 21, 2015 at 11:02 AM, Alex Morcos <morcos at gmail.com> wrote:

> Thanks for everyone's review.  These policy changes have been merged in to
> master in 6654 <https://github.com/bitcoin/bitcoin/pull/6654>, which just
> implements these limits and no mempool limiting yet.  The default ancestor
> package size limit is 900kb not 1MB.
>
> Yes I think these limits are generous, but they were designed to be as
> generous as was computationally feasible so they were unobjectionable
> (since the existing policy was no limits).  This does not preclude future
> changes to policy that would reduce these limits.
>
>
>
>
>
> On Fri, Aug 21, 2015 at 3:52 PM, Danny Thorpe <danny.thorpe at gmail.com>
> wrote:
>
>> The limits Alex proposed are generous (bordering on obscene!), but
>> dropping that down to allowing only two levels of chained unconfirmed
>> transactions is too tight.
>>
>> Use case: Brokered asset transfers may require sets of transactions with
>> a dependency tree depth of 3 to be published together. ( N seller txs, 1
>> broker bridge tx, M buyer txs )
>>
>> If the originally proposed depth limit of 100 does not provide a
>> sufficient cap on memory consumption or loop/recursion depth, a depth limit
>> of 10 would provide plenty of headroom for this 3 level use case and
>> similar patterns.
>>
>> -Danny
>>
>> On Fri, Aug 21, 2015 at 12:22 PM, Matt Corallo via bitcoin-dev <
>> bitcoin-dev at lists.linuxfoundation.org> wrote:
>>
>>> I dont see any problem with such limits. Though, hell, if you limited
>>> entire tx dependency trees (ie transactions and all required unconfirmed
>>> transactions for them) to something like 10 txn, maximum two levels
>>> deep, I also wouldnt have a problem.
>>>
>>> Matt
>>>
>>> On 08/14/15 19:33, Alex Morcos via bitcoin-dev wrote:
>>> > Hi everyone,
>>> >
>>> >
>>> > I'd like to propose a new set of requirements as a policy on when to
>>> > accept new transactions into the mempool and relay them.  This policy
>>> > would affect transactions which have as inputs other transactions which
>>> > are not yet confirmed in the blockchain.
>>> >
>>> > The motivation for this policy is 6470
>>> > <https://github.com/bitcoin/bitcoin/pull/6470> which aims to limit the
>>> > size of a mempool.  As discussed in that pull
>>> > <https://github.com/bitcoin/bitcoin/pull/6470#issuecomment-125324736>,
>>> > once the mempool is full a new transaction must be able to pay not only
>>> > for the transaction it would evict, but any dependent transactions that
>>> > would be removed from the mempool as well.  In order to make sure this
>>> > is always feasible, I'm proposing 4 new policy limits.
>>> >
>>> > All limits are command line configurable.
>>> >
>>> > The first two limits are required to make sure no chain of transactions
>>> > will be too large for the eviction code to handle:
>>> >
>>> > Max number of descendant txs : No transaction shall be accepted if it
>>> > would cause another transaction in the mempool to have too many
>>> > descendant transactions (all of which would have to be evicted if the
>>> > ancestor transaction was evicted).  Default: 1000
>>> >
>>> > Max descendant size : No transaction shall be accepted if it would
>>> cause
>>> > another transaction in the mempool to have the total size of all its
>>> > descendant transactions be too great.  Default : maxmempool / 200  =
>>> 2.5MB
>>> >
>>> > The third limit is required to make sure calculating the state required
>>> > for sorting and limiting the mempool and enforcing the first 2 limits
>>> is
>>> > computationally feasible:
>>> >
>>> > Max number of ancestor txs:  No transaction shall be accepted if it has
>>> > too many ancestor transactions which are not yet confirmed (ie, in the
>>> > mempool). Default: 100
>>> >
>>> > The fourth limit is required to maintain the pre existing policy goal
>>> > that all transactions in the mempool should be mineable in the next
>>> block.
>>> >
>>> > Max ancestor size: No transaction shall be accepted if the total size
>>> of
>>> > all its unconfirmed ancestor transactions is too large.  Default: 1MB
>>> >
>>> > (All limits include the transaction itself.)
>>> >
>>> > For reference, these limits would have affected less than 2% of
>>> > transactions entering the mempool in April or May of this year.  During
>>> > the period of 7/6 through 7/14, while the network was under stress
>>> test,
>>> > as many as 25% of the transactions would have been affected.
>>> >
>>> > The code to implement the descendant package tracking and new policy
>>> > limits can be found in 6557
>>> > <https://github.com/bitcoin/bitcoin/pull/6557> which is built off of
>>> 6470.
>>> >
>>> > Thanks,
>>> > Alex
>>> >
>>> >
>>> >
>>> > _______________________________________________
>>> > bitcoin-dev mailing list
>>> > bitcoin-dev at lists.linuxfoundation.org
>>> > https://lists.linuxfoundation.org/mailman/listinfo/bitcoin-dev
>>> >
>>> _______________________________________________
>>> bitcoin-dev mailing list
>>> bitcoin-dev at lists.linuxfoundation.org
>>> https://lists.linuxfoundation.org/mailman/listinfo/bitcoin-dev
>>>
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linuxfoundation.org/pipermail/bitcoin-dev/attachments/20151005/6e2a78a3/attachment.html>