Quote from: molecular on December 06, 2013, 11:12:07 AM

I didn't read the paper yet, but I have some problems even understanding the basic concept.

If the idea is to also accept transaction in orphaned blocks (or forks), then how can one even be sure all nodes are aware of the transactions contained in the orphans? With the current longest-chain scheme this is easy: I can just walk back through all the blocks and get a list of all transaction which is exactly the list of all valid transactions. With the proposed change I don't see how I can get a provably complete list of all valid transactions.

Please someone enlighten me.

The idea is to still accept transactions only from a single chain, but to allow blocks off the chain to also contribute confirmations to the part of the chain they reference.
Think of it this way:
Each block gives 1 confirmation to the block it references directly, and to all preceding blocks.
If there is a fork at level h, and you need to decide which one to consider as the right path to take, simply pick the block with the most confirmations (rather than the block that supports the longest chain).

RE altcoins based on this improvement:

1) I think it's more important to see a testnet that is stress-tested with large transaction loads. This is something worth doing that others have mentioned on countless occasions.

2) IMO, if crypto-currencies succeed as we hope they do (and Bitcoin is probably the most likely to grow to large sizes at this point) they will end up with high transaction rates. So we think eventually something will have to be done. Half of our paper is dedicated to what happens to Bitcoin without our protocol improvement. There is room for growth there, but we worry that after a while security becomes an issue (the 50% attack is easier, and you need to wait more blocks for the same level of certainty when facing weaker attackers).