3 ms·
So the main argument is that large blocks take longer to propagate, giving an advantage to the miners that find them. This completely ignores IBLT, which propa
by DennisP 11y ago
So the main argument is that large blocks take longer to propagate, giving an advantage to the miners that find them.
This completely ignores IBLT, which propagates a small fixed-size data structure instead. It takes advantage of the fact that miners have most of the transactions already, so sending complete copies of them is redundant.
https://gist.github.com/gavinandresen/e20c3b5a1d4b97f79ac2 https://gist.github.com/gavinandresen/e20c3b5a1d4b97f79ac2
http://www.i-programmer.info/programming/theory/4641-the-invertible-bloom-filter.html http://www.i-programmer.info/programming/theory/4641-the-inv...
- kang 11y agoBut the 'O(1) block propagation time proposal' essentially merges the consensus and networking layer. Notable problems: 1. Engineering would like to keep the networking layer and consensus layer separate because of architectural decisions for eg; difficult to push networking upgrades 2. Syncing unordered data in a trustless distributed system is a hard problem. Some propose ordering them, but that promotes censorship and centralization.
- DennisP 11y agoI don't see how it merges the layers. It's essentially a form of compression. I also don't see how deterministic ordering, perhaps based on hashing the transaction, would promote censorship or centralization.
- kang 11y agoBecause it does not work on it's own. The IBLT compression is just a way to check whether a data (transaction) is part of a dataset (transactions in mempool) when collision is low. For this to work, the mempools must be almost same. Like I said, syncing unordered trustless distributed system is a difficult problem. Suppose, engineering implements a method to do so among nodes, it will by definition be a part of the way they network but concept-wise are a part of the consensus layer.