There was this wonderful technology invented a few years ago to deal with spam. It's called Hashcash. All these hacky heuristics like block size are just dancing around the problem, and the natural solution is already present in bitcoin: smaller blocks, (down to the point of individual transactions) each mined. Don't relay things that haven't been mined. As spam or transaction levels go up, mining targets for submission go up too.

Of course this is a pretty serious redesign of bitcoin, and I'm not offering a concrete proposal at this time (but have one in the works, and I'd like to see others).

I call the parameters of these hacky heuristics "Consensus Threatening Quantities" (CTQs) because changing them induces a hard fork. Bitcoin is full of them (block time, block size, target difficulty, retarget time, etc) and bitcoin would do well to face difficult redesign questions head on, and remove them entirely. (Proposal to appear...)

On June 8, 2015 5:44:43 PM EDT, Peter Todd <pete@petertodd.org> wrote:
On Mon, Jun 08, 2015 at 02:33:54PM -0700, Raystonn . wrote:
 the attack would be expensive.
 
 For attacks being waged to destroy Bitcoin by filling all blocks with spam transactions, the attack succeeds when the attacker is well funded.  This gives well-funded private and/or public entities the means to destroy Bitcoin if they desire.  This is only true after the block size limit was implemented.  Without the block size limit, the spam doesn’t harm Bitcoin.  It simply enriches miners at the cost of the spammers, which is a nicely antifragile quality.

There will always be a blocksize limit based on technological
considerations - the network has a finite bandwidth limit.

Without a blocksize limit the attacker would just flood the network
until the bandwidth usage became so great that consensus would fail,
rendering Bitcoin both worthless, and insecure.

The worst an attacker flooding the network with transactions with a
blocksize limit can do is raise costs, without harming security. Keep in
mind, that at some point it'd be cheaper to just 51% attack the network.
Based on the current block subsidy of 25BTC/MB that's at the point where
transaction fees are 25mBTC/KB, which corresponds to <$2/tx fees - not
that cheap, but still quite affordable for a large percentage of
Bitcoin's users right now. And that's the *absolute worst-case* attack
possible.