[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[GNUnet-developers] Re: Trust vs. Reputation

From: Christian Grothoff
Subject: [GNUnet-developers] Re: Trust vs. Reputation
Date: Thu, 25 Nov 2004 12:27:14 -0500
User-agent: KMail/1.7.1

On Thursday 25 November 2004 11:59, Ludovic Courtès wrote:
> Hi,
> On Wed, Nov 24, 2004 at 02:31:40PM -0500, Christian Grothoff wrote:
> > But this should not happen with GNUnet: even if you drop out for a while,
> > you're likely to reconnect to a similar set of peers, just because of
> > your local data/hosts cache.  Also I somehow doubt that we will have
> > 100.000 (good) peers on-line at the same time anytime soon (especially
> > since we should only count peers that stay on-line for long periods
> > here).  If you have a network of say 10.000 peers and each peer has 50
> > connections (which is more what I would expect a GNUnet network of that
> > size would look like), you're also likely to earn the trust of more peers
> > at the same time (because you have more connections).  So by the time you
> > do some significant shifts in your location (which, as I said should only
> > happen very slowly anyway), you're likely to have had the chance to
> > establish some trust with a significant fraction of the network which
> > means it doesn't matter.
> I must admit that even though I like GNUnet's model very much for it is
> simple and "philosophically pleasant", I'm not that certain that their
> assumptions are wrong.  Figure 5 shows that even with a relatively low
> turnover and half of the nodes using GNUnet's model (the "private
> history" scheme), the overall degree of cooperation falls to zero when
> the system grows to between 100 and 1000 nodes only.  As nodes can't afford
> maintaining a "trust level" for every node they talked to, the trust levels
> they maintain are likely to become useless as they interact with more and
> more nodes.

Why not?  Suppose you talked to 1 Million nodes.  Each trust level is 4 bytes 
(plus say 20 to identify the node, makes 24).  That's 24 MB of data to keep. 
So you can afford maintaining a trust history for every node you talked to. 
No, that cooperation falls to zero is a different thing, their model is not 
excess based, in an excess based model you would never see that peers decide 
not to cooperate at all (since at that point all resources would become 
excess resources).

> Moreover, as GNUnet gains wider acceptance and grows, one might as well
> expect the turnover to increase I think.

Turnover may increase, but still peers will try to establish and maintain 
stable connections (that is, not change neighbours all the time).

> Also, I don't understand what makes it so likely to "reconnect to a
> similar set of peers".  If a node queries data which the peers it is
> connected to don't have, it will _have to_ talk to other peers it may
> not already know, won't it?

Nope, that's again one of the key differences in the P2P model: GNUnet uses 
anonymous routing (traffic is indirected) and not a Direct-Connect/gnutella 
style overlay where in the end a direct end-to-end connection is necessary. 
This is one of the reasons why there is more locality in GNUnet and why the 
economy will work better for GNUnet then for some 'arbitrary' P2P design.


reply via email to

[Prev in Thread] Current Thread [Next in Thread]