Fabio Pietrosanti (naif):
Il 9/27/14, 2:33 AM, Mike Perry ha scritto:
We could also handle controlled rollouts to fractions of their userbase to test the waters, and slowly add high capacity nodes to the network to support these new users, to ensure we have the people ready to accept payment for running the servers, and maintain diversity.
I read your very detailed estimations and improvement paths, i love it!
However i see that the main suggestion to increase the "network capacity" can be simplified as follow:
- improve big nodes ability to push even more traffic
- add more big nodes
Other improvements are to reduce the "consensus size" and "directory load", but not specifically on network capacity.
While this is the obvious way to "add more capacity" i feel that's going to have impacts such as:
- reduce the "diversity" (thus the anonymity, because few players will
handle most of the network's traffic) 2) make it "irrelevant" for anyone to run their own small/volounteer relay
That sounds like the "easier way" to scale up in a defined amount of time and with a defined budget, but imho also with consequences and pre-defined limits.
I feel that the only way to scale-up without limits and consequences is to have end-users became "active elements" of the network, where we have success story such as Skype.
End-users have important network resources available that can be estimated and used (with care).
Not all end-users are equal, i'm now on a 2M Hyperlan line (damn digital divide!), but someone else in Stockholm or San Francisco it's on a 1000M/100M fiber connection @home (not in a datacenter) and while in Milan i've a 100M/10M fiber!
That bandwith resources are amazing, usually quite cheap (home broadband lines), widely available in the end-users hands.
IMHO those are the bandwidth resources, widely available, cheap, very diverse/sparse that could help the Tor network to scale-up.
How to use it properly within/for the Tor network? That's a different topic.
It's the same topic: I'm arguing that we want to use the 100M fiber connection, and maybe the 10M connection, but definitely not the ADSL link with only 256kbit upstream. The latter costs more bytes to tell clients about than it contributes to the network.
We can cut these ADSL relays from the network and turn them into bridges using the bandwidth authorties. Or have the default relay mode be to start as a bridge and get promoted to a relay once you are measured.
As for diversity, we can better achieve diversity through proper network allocation based on the current node selection algorithms and load balancing, so we actually know that our desired percentage of traffic is going through the geography/jurisdictions/organizations we want.
Keeping thousands of junk nodes that only carry a tiny fraction of the Tor network capacity just so we can pretend we have diveristy is no solution. It's wishful ostrich thinking.
Slow/junk home nodes also have worse mix properties than fast nodes, due to less concurrent traffic running through them. They are thus more useful to surveil externally for correlation, and probably also easier to compromise.