[Cryptography] Almost decentralized currency

L. M. Goodman lmgoodman at hushmail.com
Tue Jun 24 07:25:17 EDT 2014



On 6/24/2014 at 6:15 AM, "ianG" <iang at iang.org> wrote:

>
>
>Yes.  If you are dealing in networks of smaller internally trusted
>groups then you maybe don't need the blockchain at all, because 
>they
>already trust members internally.  Use that trust, run a small
>centralised system, locally.  Then, your challenge is to bring the
>groups together.  What do they want?  What does relating to other 
>groups
>mean?

This is not really what I'm describing. There is no network of smaller internally trusted groups. The paper I'm linking to detects Sybil node among a single, global network. 

Instead of expanding hashing power to be part of the consensus group, nodes need to establish trusted peering relationships with honest nodes in the network. If most people aren't colluding to destroy the network, you can maintain a singleton.

You do need a blockchain, for the same reason that a business needs double entry accounting. It's an added security that makes cheating much easier to detect.

To be clear, I'm not describing small networks of people who strongly trust each other. The trust relationships established are very weak. The meaning of a link is really "yeah, I don't think you're a Sybil node", not "here are the keys to my house".

>
>BTW, a lot of this is not about crypto.  It's far more about
>understanding what the groups are and do, and molding the crypto to
>assist that functionality.  You can't just assume the groups 
>behave like
>some paper on Sybil attacks and then go find that plastic world.


The paper was tested empirically on real world datasets. The fast-mixing assumption seems to hold, even among networks presenting internal clusters.

But yes, any such system does not depend on crypto but on human assumptions. The assumption that a real world social network is fast-mixing seems more realistic to me than the assumption that bitcoin miners will not collude.



More information about the cryptography mailing list