[mnet-devel] Hack day: my plan

[Some Guy]

 --- Zooko O'Whielacronx <zooko at zooko.com> wrote: 
> <my address please hide> asked a bunch of questions of icepick, but I'm replying with my own
answers.
They were meant for all.  Wow you've certainly given me lots to read.  Thanks!!

> > Chord eh.  I that your favorite DHT topology?
> 
> I don't have a favorite at the moment.  Chord is an excellent choice for 
> experimentation.
Hey if it works, it's all good.

> > I have one gripe about them: since request only go
> > one way around the ring you cann't know if the guy that's making you do work is doing work for
> > anyone, since he isn't doing any for you.
> 
> Yes, that's an interesting issue.  "Bidirectionality", as the Kademlia paper calls it.
Thanks for the word.  I'll read thier stuff I.  I guess I haven't yet.

> > I've got this strange picture in my head of a sloppy
> > two way Chord where nodes maintain have random connections distributed to the other nodes in
> the
> > ring based on 1/<hash distance>.
> 
> So the idea of a "flexible" Chord with links existing in a certain 
> distribution rather than being fixed at specific intervals is separable from 
> bidirectionality, isn't it?
> 
> One could have normal traditional inflexible Chord, but just change it so that 
> you can route in either direction.  Then it would be bidirectional Chord.  Did 
> I miss anything?
Maybe you're right.  I thought part of the big deal behind Chord is that you can't overshoot. 
With a two way chord you don't have that problem, but I guess you're right.

> Here is a paper about a flexible Chordish-like thing.  It requires the links 
> to conform to a certain probability distribution:
> 
> "Symphony: Distributed Hashing In A Small World"
> http://citeseer.nj.nec.com/568921.html
I heard of that.  Thanks I'll read it.

> and here is a paper which infuriatingly mentions that a flexible variant of 
> Chord would work fine without giving any citation or explanation of how it 
> would work.  It has a footnote that says something like "P2P researchers all 
> agree that a flexible variant of Chord would work fine.".
Here's sort of a sloppy proof for the sloppy network:
Picture the right where the probablity I have a connection to you is based on 1/<distance>.  So
I'm connected the next couple nodes on each side.  Then the connections get sparser as go around.
Or think of it the otherway.
I have 10 connections to random places on the ring.  I have 10 to my 10th of the ring.  I have 10
to my 100th. ......
It seems logical this thing will route in about log10(N).


> "The Impact of DHT Routing Geometry on Resilience and Proximity"
> http://www.cs.washington.edu/homes/gummadi/papers/p1101-gummadi.pdf
Once again thanks, more stuff to read. :-)
 
> > Getting a fixed topology where the bad guys are spread evenly throughout is a big goal for
> > anonymity and for DHTs, since cancer nodes can otherwise wipe out specific material.
> 
> I don't really like that kind of approach.  It doesn't seem very feasible, and 
> I think that a friend-net kind of approach should be much safer as well as 
> more feasible.
I'm exploring both the friend-net idea, but also this stuff.  Remember a friend net may shoot you
in the foot since traffic analisis reviels who's friends with whom.

> > I think I've got a way to make a network do this.  Has anyone seen any other good projects
> that
> > try to do this?  Or does anyone have any good ideas.
> 
> I'm interested in your idea!
Check this out:
Divide nodes up in a d dimensional sloppy CAN, with a g^d cells.
Every iteration do reorder one dimension.  So for example 
NewX= hash(newXsalt,nodeID)
All nodes stay in contact with nodes which are different in just one dimension.
When a new iteration shifts one dimension around and I get a new neighbor, I'll have d-1 cells
with witnesses which should be able vouch for him.

I think I can prove:
A) a randomly distributed adversary will always stay randomly distributed.
B) a clump of evil nodes will get broken up or disconnected from the net.

I've also came up with a cool Hash Bank idea, which would force each node to store a few GB of
data to keep the population of evil nodes limited.  But that's another story.

The whole basic idea is that not just anyone can voach for an identity.  Everyone is assigned a
few bosses.  

Seems like something beurocracy, politics, and economies maybe could use somehow.  "who watches
the watchers?" "We do."  Ok maybe that's getting a little out there.  Some people do think though
that P2P systems might give us ideas for our social systems.

> Hack Day is on Sunday, so our Jewish hackers can participate.  I don't know if 
> any of our Muslim hackers will participate in this one, since it is Ramadan.
It think they can.  They just can't eat.

Once again thanks for all the papers.  I won't run out of stuff to read this weekend.

__________________________________________________________________

Gesendet von Yahoo! Mail - http://mail.yahoo.de
Logos und Klingeltöne fürs Handy bei http://sms.yahoo.de


-------------------------------------------------------
This SF.net email is sponsored by: SF.net Giveback Program.
Does SourceForge.net help you be more productive?  Does it
help you create better code?   SHARE THE LOVE, and help us help
YOU!  Click Here: http://sourceforge.net/donate/
_______________________________________________
mnet-devel mailing list
mnet-devel at lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/mnet-devel