[mnet-devel] progress implementing emergent networks

[Zooko]

Myers reported (I think) that his twisted Chord network passes a unit test 
where you start with two separate Chord networks and introduce one node from 
the first net to one node from the other, and then the two nets merge.

Meanwhile, I've finished implementing a version of ent (based on Kademlia) 
which keeps only one node per k-bucket, and fixed several bugs, but there 
remains some bug that I haven't investigated (I'm out of time) which causes it 
to fail the basic "contruct a network, publish a block, fetch the block" unit 
test.  (Please, someone fix it, as I'm probably busy this week.)

I just wanted to comment that there is no way known (to me at least) for 
Kademlia to pass the unit test that Myers is using on his Chord net -- merging 
two separate nets into one.  Kademlia can't do that AFAIK.  (This is one way 
of observing the "Kademlia doesn't self-heal" problem.)

I also wanted to mention that Chord can sometimes fail, too, if the nets 
happen to line up so that the resulting merged Chord net is "loopy".  The 
Liben-Nowell paper [1] explains how to fix that.

--Z

[1] http://citeseer.nj.nec.com/553810.html


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
mnet-devel mailing list
mnet-devel at lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/mnet-devel