For additional information: https://blog.torproject.org/category/tags/entry-guards

Yes, I have been considering i2p, the 'garlic router' which seems to have hedged against a lot of attacks that could be possible on tor.

I agree, I2P is interesting. I don't know why it doesn't get more press or support. According to the website some reps of the I2P team will be at 30C3 in 2 weeks time btw.

I have a theory,

There are 7 directory authorities in Tor, if they don't function then the tor network is dead. So that is just seven people you would need to abduct and torture (to take over the tor network) or 7 drone strikes to kill all tor traffic. I am unsure but I think this is not the case for i2p.

EDIT: So that might be a reason for NSA to support Tor over i2p.

No one uses I2P because it hasn't been deeply studied. No one studies it because no one uses it.

Given the need for critical mass, is there any way some or all of the advantages of i2p could be retrofitted to tor, or are the differences more fundamental?

I am not sure but I think the differences are too fundamental. But it would probably benefit anonymity to merge multiple projects.

isn't i2p simply tor but with 2 hops instead of 3?

You can select your own circuit length.

How? The tech intro on the website describe tunnels (https://www.i2p2.de/techintro.html) with 3 parts.

#1 is Alice the sender or Bob the receiver

#2 is a participant

#3 endpoint gateway

Two such tunnels are used when Alice and bob communicate, for a total of 4 nodes for the communication between alice and bob.

With tor hidden services, it is the exact same scheme, but with 4 nodes per tunnel rather than 3. (or 3 nodes rather than 2 if you do not want to count alice/bob as nodes). In total, 6 nodes between is used for the alice and bob communication. (https://www.torproject.org/docs/hidden-services.html.en)

Please explain how this is incorrect. Is the tunnels as describe the I2P documentation just illustrations of how an tunnel might look like, but isn't actually how it is in practice? How does it solve the problem that tor fixes with guard nodes (as this is the context in the above comments and article)?

I am much interesting in I2P, but the above details/questions has held me back.

You can have more than 1 participating hop

Let's say we have Alice and Bob, alice wants to have 4 hops both outbound and inbound and Bob wants to have a 2 hop inbound and a 1 hop outbound

Alice's tunnels would be:

Alice -> Participant A1 -> Participant A2 -> Participant A3 -> Outbound Endpoint 1 -> ?

Alice <- Participant A4 <- Participant A5 <- Participant A6 <- Inbound Gateway 1 <- ?

Bob's tunnels would be:

Bob -> Outbound Gateway 2 -> ?

Bob <- Participant B1 <- Inbound Gateway 2 <- ?

For Alice to send to Bob and receive a reply the round trip path of the message would take would be

Alice -> PA1 -> PA2 -> PA3 -> OBEP1 -> IBGW2 -> PB1 -> Bob -> OBEP2 -> IBGW1 -> PA6 -> PA5 -> PA4 -> Alice

Since the above techintro documentation seems contradicting, I looked around and found a better page at http://www.i2p2.de/how_tunnelrouting.html which explains what you said. Tunnels got a max length of 7, and is clearly easy configurable in a session config file.

Thanks, Im going to take second look now that that is cleared up.

The client always selects the route on any onion routing network. They have to, because only they know the route. It's been a while since I played with I2P, but i'm 95% certain you can put push the circuit length up