Re: [Help-gnunet] Paper: "Request Algorithms in Freenet-style Peer-to-Peer Systems

[Igor Wronsky]

On Tue, 18 Mar 2003, Doug Bostrom wrote:

> I've been following with interest the development of "Freenet" for some
> time. While perusing the developer mailing list there, I ran across this
> paper referenced in a posting:
> http://www.lclark.edu/~ppair/mache.pdf
> Apologies for posting this here, but I'm not a developer and thus did not
> want to join the developer mailing list. I've only just discovered Gnu-net,
> and thought that since your network is in an earlier stage of deployment
> than Freenet it's possible the ideas cited in this paper might be easier to
> adopt should they prove valid.

I've read the paper now - not that I'd have much to comment or
feel like a master of routing business, but atleast you won't feel
totally ignored if I say something. ;D

It seems that some of the methods in the paper would
compromise anonymity. Each of them seems uncompatible with
current gnunet protocol. The reason is that in gnunet, node
contact information is not passed in requests or replies,
but by occasional HELO messages which are (as far as I know)
always accepted if correct and redistributed in a rather
random fashion. Unsuccessful queries in gnunet are not
answered at all.

Keeping HELO of each operational node for its validity time -
and trafficcing in them - might be eventually infeasible (this
does not however mean that we'd be connected to each of them
at once), but this is far from being a problem at the moment.
Besides the gnunet economy is used for preferring who to talk
to when it gets crowded, and can be extended to HELO business
if not already done so.

I don't claim that the current way of doing things in gnunet
is the best possible. I don't claim that the ideas of the
paper weren't good. Its just that we'd need (if only a simulated)
comparison of what gnunet does and what freenet does (with or
without the proposed algorithms), to really be able to select
and/or rank these methods. The only thing I'm fairly certain
of is that if developed well, any method that chooses its
behaviour based on how it likes the other nodes (as gnunet tries
to do) should be more practical than methods that treat every
node as equal (as freenet did when I last time looked). The
reasoning is that not every node has equal bandwidth, storage
space, contents, or will to use its resources to help others. It
very well might be a better idea to find 10 "good guys" from
a million nodes and prefer them to talk with than to get avg
route length of 3 to a million no-good dial ups that pop in
and out uncontrollably and are capable of serving/routing
practically nothing.


Igor