[Author Prev][Author Next][Thread Prev][Thread Next][Author Index][Thread Index]

Re: Optimal ratio of node types



On Wed, Oct 04, 2006 at 02:12:23PM -0400, Tim McCormack wrote:
> Anybody know of research that has been done on the optimal ratio of node
> types in a Tor-type OR network?

Assuming all traffic goes through three hops (and for the most part it
does; the four-hop cases are a bit weirder and a lot less frequent)
then, to a first approximation, we want at least 1/3 of capacity in
exit nodes.

(That's capacity, btw, not number.  2 exit nodes that do 10kbps are
not really much better from a capacity POV than 1 exit node that does
20kbps.)

This is, of course, an approximation; the real figure will probably
be a bit different because of many issues, such as varying exit
policies, socket limitations, differences between the cost of handling
encrypted OR traffic and handling traffic to the network, and so on.

It doesn't go the other way around, btw: since all exits can be
relays, it doesn't hurt to have an all-exit network.

There's also directory caches.  More of those could never hurt,
especially on middleman nodes.

By the way, this kind of came up in a recent change we made that
should show up in 0.1.2.2-alpha: the new code avoids using exits as
relays nodes when there are lots more non-exits than exits.

yrs,
-- 
Nick Mathewson

Attachment: pgp0y5DjOKoKA.pgp
Description: PGP signature