Are individual circuits exposed in inter-relay traffic?

Question

There are thousands of Tor relays. And so it's arguably impractical for each relay to maintain persistent connections with all of the others, even without padding traffic.

At the other extreme, a given relay would only connect with another when it received a cell that was addressed there. Is this how Tor currently works?

I gather that relays perform no explicit mixing. But, beyond padding to constant cell size, are there steps to obfuscate inter-relay traffic? For example, can obfsproxy be used between relays?

Is the cell structure of individual circuits obvious in inter-relay traffic? To what extent is it obscured through TCP mechanics?

Answer 1

Links between Tor nodes are encrypted using TLS (on top of the telescoping onion routing circuit encryption) so the number of cells corresponding to each circuit is obscured from an external adversary. TLS also authenticates the link between nodes, to resist active attacks. It's not clear yet how to protect links better; padding might help but so far all the proposals which offer good security have prohibitively high overhead in terms of extra traffic and extra latency.