[p2p-hackers] Non-transitivity in DHTs

[coderman]

On 10/28/05, Michael J. Freedman <mfreed at cs.nyu.edu> wrote:
> ...
> Beyond communicating these details to the development community, we had
> hoped to engender further discussion on how other DHT designers and
> implementors have dealt with similar problems and on what techniques
> should be used to avoid such limitations in the future.

i've found a number of the ad-hoc routing protocol / wireless mesh
routing protocols[1] to be useful with regards to non-transitivity and
decentralized networking in general.  at the application level we are
comfortable with easy promises of flow controlled streams between
peers; lower level routing protocols get no such easy service.

for example, some protocols utilize unidirectional links for
distribution of route state even though such a link is not useful for
transport (lacking bidirectional communication).  the wireless
protocols offer interesting uses of a true broadcast transport that
are particularly relevant to fully decentralized peer networks.

the real crux of the problem is reputation (and strong identity to
bind reputation to, regarding the previous thread).  for this reason i
think DHT's will work best using smaller groups of peers which are
known rather than larger groups of unknown and untrustworthy peers.
[for example, coordinated attacks against DHT state using multiple
malicious nodes; it is extremely hard to protect against this in a
DHT]

[1] http://en.wikipedia.org/wiki/Ad_hoc_protocol_list