[Bitcoin-development] Fee drop
pete at petertodd.org
Fri Feb 28 11:18:26 UTC 2014
On Tue, Feb 25, 2014 at 10:09:23AM -0800, Jeremy Spilman wrote:
> If I understand correctly, the risk here is this would open a
> historically large discrepancy between MIN_RELAY and the expected
> minimum fee to actually obtain block inclusion. I don't know if
> that's true, but I think that's what Peter is saying makes it
> different this time.
That's exactly the problem.
Of course every time we make a new transaction type standard we also run
that risk, but at least it's a temporary situation and we can expect to
get hashing power on-board fairly quickly. With such a low MIN_RELAY
that's not true, and in an absolute sense, the funds required to DoS
attack the network are fairly low.
> On Tue, 25 Feb 2014 08:55:58 -0800, Mike Hearn <mike at plan99.net> wrote:
> >Nodes that are encountering memory pressure can increase their min
> >relay fee locally until their usage fits inside their resources.
> >It's annoying to do this >by hand but by no means infeasible.
> Perhaps this is just another way to think of the floating fee
> problem. What does MIN_RELAY need to be so that my local resources
> stay within some reasonable limit (and 'reasonable' means different
> things to different nodes).
> We have an input gate on transactions entering mempool, we persist
> mempool, and I don't know the specifics but, I assume there's some
> expiration policy other than block inclusion to clear out a Tx from
> mempool. But are transactions prioritized in any way after they make
> it into mempool today?
There's currently no expiration policy at all; that's the root of the
DoS problem I was referring too.
> How closely should mempool selection align with the expected block
> inclusion? I think if they align perfectly in theory that means
> optimal mempool resource allocation. For example, a miner would push
> out cheaper transactions which they were previously hashing against
> to make room for higher fee transactions (bsaed on max block size or
> orphan rate projections), but do we do the same for mempool? E.g.
> - After hitting X number of transactions, the fee has to be larger
> than a transaction in mempool in order to get in,
> - That in turn that ejects a random transaction which paid less
> fees than the incoming Tx from mempool
> - We would have to consider how ejection would work with chains of
> unconfirmed transactions (cumulative average fee/kb?) but again in
> this case, you would want to 'do what miners would do' if you could
Have you seen the mempool superblock design that keeps getting
suggested? jgarzik has the most recent write-up here:
I was working on a relatively ambitious version of it last summer that
calculated the fee/KB for transactions, including depedencies, and then
simply ordered the mempool with highest fee/KB first. The idea was you
could then easily limit the total size of the mempool and drop
transactions with the lowest fee/KB first. Transactions that paid less
than the lowest fee/KB in a max-size mempool simply would not get
relayed at all. Pity had to put it off for higher-priority work.
What's interesting is how this makes zero-conf transactions even less
safe: all you have to do to double-spend one (or more!) that pay X
fee/KB is broadcast enough transactions paying X+e fee/KB to push out
the unconfirmed tx from mepools around the network, then broadcast your
double-spend. Obviously the economics of this are going to make attacks
frequently profitable, especially if you can attack multiple targets at
once. You can of course have schemes where you don't entirely drop
transactions, saving, say, the inputs they spend and a transaction id,
(so a rebroadcast can succeed) but that just reduces the effectiveness
of the attack by a constant factor and makes it possible to get into
complex situations where your funds are locked and unspendable.
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 685 bytes
Desc: Digital signature
More information about the bitcoin-dev