On Mon, Jun 08, 2015 at 02:14:01PM -0700, Raystonn . wrote:
> > there is no memory pool cap currently
> 
> Real hardware does not have an infinite amount of RAM.  Memory pool sizes 
> cannot grow unbounded.  Some transactions with insufficient fees do get 
> dropped today after many hours.

Actually they don't, which is an unfortunate problem with the existing
mempool implementation; the only way a transaction can be removed from a
Bitcoin Core mempool is through it getting mined, double-spent, or the
node restarting.

The protection that we have against that attack is that you need access
to a lot of bitcoins to pay enough fees. With the 0.01mBTC/KB minimum
relay fee and $230 USD/BTC that works out to about $2.3kUSD/GB of ram
consumed, and furthermore, actually getting that many transactions to
propagate over the network is non-trivial. (no, I'm not going to tell
you how)

The obvious solution is to cap the size of the mempool and evict
transactions lowest fee/KB first, but if you do that they you (further)
break zeroconf security. On the other hand, if you don't break zeroconf
security an attacker can prevent reasonable fee transactions from
propagating.

I probably should get around to fixing this...

-- 
'peter'[:-1]@petertodd.org
0000000000000000127ab1d576dc851f374424f1269c4700ccaba2c42d97e778