Ian Clarke wrote:
On 3/12/06, Serguei Osokine <[EMAIL PROTECTED]> wrote:
By the way, as a practical matter, does Dijjer grow its routing
tables with network size?
I will take this as Oskar isn't really familiar with the details of
Dijjer (I think he is much more interested in the mathematics rather
than the practicalities of implementation). Dijjer does not scale its
routing tables with the log of the network size, basically it tries to
make the routing tables as big as possible within a given constraint
(I think the default is 20 connections). While this means that it
doesn't have log(N) scalability, in practice it doesn't really make
much difference and it means that we don't need to try to calculate
the total network size.
To put actual numbers to this, I did simulations of point to point
routes using the same algorithm that I think is being used in Dijjer
(the one described in my thesis, with 20 shortcuts, but without using
"local" links):
1000 4.4446
2000 5.066
4000 5.7604
8000 6.546755
16000 7.476
32000 8.34637
64000 9.434343
128000 10.747675
256000 12.572815
512000 15.743066
Success-rates were above 99% for all levels. These are just the lengths
of routes, and do not take things like caching into account (off hand,
you could probably see each cached document as an extra shortcut, since
having the document required is the same thing as having an edge to its
"home").
So at least up to 150,000 nodes or so you should have no problems with
20 neighbors. Above that one might want to consider having more.
// oskar
_______________________________________________
p2p-hackers mailing list
p2p-hackers@zgp.org
http://zgp.org/mailman/listinfo/p2p-hackers
_______________________________________________
Here is a web page listing P2P Conferences:
http://www.neurogrid.net/twiki/bin/view/Main/PeerToPeerConferences
