[Cryptography] Finally! Hyperledger is a "trust N out of a selected M" ledger system!

[L. M. Goodman]

On 7/10/2014 at 7:04 PM, "Lodewijk andré de la porte" <l at odewijk.nl> wrote:
>

>Problem is that consortia are never good enough. There's always 
>too big an
>opponent that can take down too much of a consortium. Bitcoin is a 
>tease
>stronger than that. But much more expensive.
>
>I don't think it will take off though, there doesn't seem to be an 
>early
>adopter advantage.
>
>Thoughts?
-------------------------
A)

It's great that this exists. Like you're saying, I think it may have a hard time taking off because of the explicit hierarchical nature. People (in the west) seem to strongly prefer system which aren't explicitly hierarchical, even though they can implicitly be. I think that's a hunter gatherer egalitarian instinct kicking in.

It's sometimes confusing to talk about "decentralized" systems, because most of the time, people mean "non concentrated", or "non hierarchical". Decentralized only means that there is no explicit hierarchy. A good analogy is embryogenesis: it's a fully decentralized process because all cells run the same DNA, but it self arranges into something hierarchical.

Likewise, no one in Bitcoin is given an explicit role a priori, but a hierarchy of miners spontaneously forms. The case of Bitcoin shows that the emergent hierarchy can end up being far more concentrated than an explicitly chosen one. Yet, right or wrong, there's a strong preference for non explicit hierarchies.

-----------------------
B)

As I've mentioned before, I think the gap can be bridged by using social networks to form consensus... but it's tricky. I've been playing with some datasets from G+ and Facebook (snap.standford.edu) but they were too incomplete to be useful

I model consensus on a social graph by assuming that everyone starts by picking a value between 0 and 1. In order to reach a consensus, users chain  two types of operation:

 - Diffusion: a node's value changes to the average of its neighbors
 - Non-linear transform: a node's value is replaced with it's image through a sigmoid function, stretching the value towards 0 or 1

For instance, a model could be: do 10 diffusion steps, and then pass through a hard threshold  (fun x -> if x < 0.5 then 0 else 1)
This model is exactly equivalent to this strategy: "set your bit to the bit you'd be most likely to encounter when doing a 10 step random walk in your neighborhood"

Softer sigmoid functions and different schedules can be played with. This has the potential to be resilient to a Sybil attack, because it is difficult for Sybil nodes to attach themselves to the graph. Specifically, the cut set between the Sybil nodes and honest nodes will be small. Thus, a short random walk starting from an honest node is unlikely to be absorbed into a cluster of Sybil nodes.

In the ends, it boils down to the spectrum of the stochastic transition matrix of the social graph (call it G). If the restriction of G to the subspace of honest nodes has small eigenvalues, it means the graph isn't too clustered and exhibits small world effect. Convergence is rapid and it's easy to limit the influence of Sybil nodes. If the eigenvalues are high however (the graph has large clusters), then you need to be far less aggressive in the schedule to ensure convergence, but that also increases the influence of Sybil nodes.