On Sat, Dec 17, 2011 at 8:28 PM, Gregory Maxwell <gmaxwell@gmail.com> wrote:

On Sat, Dec 17, 2011 at 8:37 AM, Christian Decker
<decker.christian@gmail.com> wrote:
> My idea was to structure the network in a hypercube and use prefixes to
> address different parts of the network, and use those prefixes also to find
> the location where an item (transaction, block, ...) should be stored. Each
> vertex in the hypercube is a small, highly connected, cluster of nodes.

I strongly advise people who are not me to use this sort of scheme, so
that I may enjoy the benefits of robbing you blind.

.... But really, saying "some sort of DHT" without basically
presenting a working implementation that demonstrates the feasibility
of solving the very difficulty attack resistance problems these
schemes have basically triggers my time-wasting-idiot filter. (Or
likewise, presenting a fixed network structure that would have a nice
small and easily identifiable min-cut...)

I don't doubt I'm completely alone in this, though perhaps I'm more
of a jerk about it. Even if your actual proposal might have some
merit you should be aware that every fool who has operated a
bittorrent client has heard of "DHT" and, although they may not even
understand what a hash table is, many have no reservation going around
suggesting them for _every_ distributed systems problem. Want to scale
matrix multiples? DHT! Want to validate bitcoin blocks? DHT! Network
syncup slow (because It's bound on validation related local IO)? DHT!
I suggest people solve the real problems first, then worry what name
to give the solutions. ;)

To address gavin's tragedy of the commons concern, one useful feature
would being able to mutually authenticate a peer... then full nodes
could pick and choose which lite nodes they're willing to do (a lot
of) hard work for. This would also be valuable because some modes of
lite operation require non-zero trust of the full node being queried.