Excellent, thanks! In summary (obviously there is substantial overlap):
1. Anonymous participation is the idea that you can participate in a
distributed system without revealing that to third parties, or at least
to third parties that you do not trust. Think of the structure of
underground networks, where members only communicate with people they
really trust, because every contact that you don't know about could work
for the Gestapo, or the KGB, or the RIAA. This implies serious
restrictions on the topology of the graph, which means different message
routing protocols than classic DHT like Kademlia. It also require secure
protocols to add/remove contacts, or to connect/reconnect to the graph.
Freenet's "dark net" does lots of that, and the question is, what would
it take to make it mainstream?
-- Christian Huitema, also mentioned by Michael Rogers
2. Hidden rendezvous is the idea of enabling something like Skype on a
DHT, but hiding who communicates with whom. Suppose that I connect to
the network anonymously, e.g. using a public Wi-Fi access point, and
that you do the same. How do we find each other without revealing our
location to the NSA and its peers? One potential solution is that we
agree on a secret and derive from it a series of random numbers, say
hash(secret, time-of-day, my-name). Then use that as a key to publish an
IP address in the DHT. A bit clumsy, of course. Can you do better?
-- Christian Huitema
3. Adversary Model and Incentives.
I think the standard for most decentralized systems including DHT's and
consensus networks is in the "honest" vs "Byzantine" model, with no
regards to incentives. I think the incentive model is a recent thing
since Bitcoin, and it isn't well accepted.
In the incentive model, you might consider three groups of actors.
"Honest", "rational", and "byzantine". The rational my diverge from the
prescribed protocol for self gain, but the honest deviate.
-- Jae Kwon
I wonder what the standard adversary model is for DHTs, and what happens
when some significant fraction of participants are malicious.
-- Greg Troxel
Dealing with Sybil Attack and Eclipse attack (Many corrupt nodes
choosing DHT identities close to some value).
-- real, Micheal Rogers
Creating a rigorous adversarial model for DHTs. (I think that we still
don't have one. Most articles on this subject resort to experimentation
because they can't prove correctness).
-- real
4. Practical limits.
Running a DHT despite the NAT problems. (While many present non-elegant
technical ideas to route around NATs, there might be a nice theoretical
solution).
-- real.
Navigation using Virtual DHTs (Like done in Cjdns. I think nobody really
knows to prove why it works, and whether it is going to scale).
-- real.
5. Applications.
Go back to square one and ask what other pointer-based data structures
would lend themselves to distributed implementation...
-- Micheal Rogers
For example, IPFS, essentially a singly-rooted git tree distributed via
bittorrent. Something that this makes possible, is everyone publishing
trees and software intelligently combining those. ... My idea is a piece
of data is available through many paths. For example, I want
.../book/by/Frederik Pohl/Gateway/ and .../book/award/Hugo/1978/winner/
to point to the same location. Navigation could be, in part, winnowing
of possible completion paths. It could be coupled with a signing system
that allows individuals or organizations to authoritatively publish
nodes. So, different groups could curate different subtrees. All of
these are then conglomerated so the ideal content is retrieved when a
user requests something.
-- Will Holcomb
On 2/01/2015:
Request for comments: what are the open topics in DHTs?
I'm asking on behalf of a 4th year CS student preparing to do the final
year project; with possible extension into Masters. Having worked with
DHTs, the bug appears to have struck...
_______________________________________________
p2p-hackers mailing list
[email protected]
http://lists.zooko.com/mailman/listinfo/p2p-hackers
