CRDT & blockchain comments

[gritzko]

Migration of this thread:
https://twitter.com/juanbenet/status/748991617495859201

[jbenet]

Thanks @gritzko -- i will read the thread more and write out my thoughts. But more importantly, i will read your paper. It sounds very exciting 😄

[AFDudley]

I gave this a quick once over, this doesn't provide Byzantine Agreement. Nor does the punishment model provide any sort of Sybil resistance. This model was interesting to think about, but it doesn't seem to provide any protection in an adversarial environment.

[gritzko]

@AFDudley that is exactly the point.

This is exactly the point with everything eventually consistent, e.g. Amazon Dynamo: it is already good enough. Making it perfect is so insanely expensive that it makes it actually much worse in price/performance terms.

The idea is: make coin splits a possibility which is less than the rounding error. There are much more significant real-world risk factors which can not be physically eliminated anyway.

[AFDudley]

If the point of this is to run it in a federated context, that's cool it's a lot more interesting in that regard, I still don't like the blame model though.

[ebuchman]

I like the approach, but the challenge is to assure economic security against double spends, especially given a globally active network adversary. Pushing all the validation to the edges is reminiscent of what Eg Peter Todd has been advocating for in a scalability context, but it remains to be seen how the economic security guarantees can be preserved. Perhaps some mix of an atomic broadcast backbone which can manage and slash security deposits combined with this sort of coin-consistency-only-at-the-edges could work. Punishment in IP space alone (ie evicting peers) is almost certainly not enuf to secure a real coin.

Entanglement matrix only seems to work if peers reliably sign everything and pass it on . But byzantine peers can't be relied on to do that.

[gritzko]

Sorry for slow responses, I am vacationing right now.

I mentioned Hawala as an inspiration for the proposal. Hawala is honor-based: you have your life's business and income (and your children are likely to inherit that), iff you behave well.
Similarly, Swarm coin assumes that peers have some real-world well-established reputations (there is a distinction of "peers" and "clients" )
It definitely does not work with random peers "from the internet".
Even BitTorrent had to introduce some short-term reputations (tit-for-tat) to make the network work.

Mandatory signing is a purely technical thing, it can be mandated at the protocol level (ops are not accepted unless signed).

Thanks for the questions, I do my best to extend the docs based on the feedback.

[gritzko]

@ebuchman I added a discussion on Byzantine vs Entanglement

[gritzko]

@AFDudley I added a discussion on the general approach https://gritzko.gitbooks.io/swarm-the-protocol/content/coin.html#the-general-approach
Regarding the "adversarial environment" comment, I assume that (1) peers are long-lived and established/reputable and (2) double-spend it too easy to detect. So, it is not about random peers "from the internet". I will provide more detail in peerage.md, but that will take a day or two.

[ukstv]

@gritzko Quite surprised IOTA (https://www.iotatoken.com) never popped in the discussion. Their approach look extremely similar to what Swarm split-coin proposes.

@AFDudley Could your comment on peer identity and sybil resistance be applied to IOTA as well, by the way?

[gritzko]

@ukstv It popped in another discussion I recently had. IOTA is DAG-based, but it relies on "truth as the longest history branch", i.e. implicit voting. "Swarm coin" has no voting. Also, S.Popov is a mathematician, so their whitepaper is rather light on logistical details. That's all I can say at the moment.