Nothing has any defence against global traffic analysis.

Tor still does better than anything else. I don't think it's worth scaring people away from using Tor, because whatever else they'd be using instead is certainly worse.

EDIT: Parent comment originally said words to the effect of "Reminder that Tor has no defence against global traffic analysis".

> Nothing has any defence against global traffic analysis.

It's possible to guard against global traffic analysis by establishing permanent fixed-bandwidth links between each node and sending traffic along them even when they aren't assigned to a circuit (or the circuit is idle). Then there is nothing to passively analyze because the amount of traffic between each node is always the same.

The problem is that this consumes a very large amount of bandwidth.

This works against global passive adversaries.

If you instead ask yourself "how many Tor nodes are out there"[1], and then ask yourself "what is the NSA's annual budget"[2], the concept of a global active adversary makes me a bit nervous.

[1]: http://torstatus.blutmagie.de [2]: https://www.theverge.com/2013/8/29/4672414/leaked-snowden-do...

The scary thing about a global passive adversary is that the attack is undetectable. Active attacks are harder to defend against but they're also harder to keep secret.

You may want to investigate i2p which is resistant to traffic analysis.

And does not run 100%bandwidth all day and all night.

I think you're referring to traffic analysis for protocol identification, which is something else entirely, and Tor handles that as well. Resisting protocol identification can be done very efficiently -- you basically get the same information theoretical efficiency as the protocol you're mimicking does, and the primary failure mode is to not be mimicking it completely accurately. It's theoretically possible to mimic the target protocol perfectly, and each time someone finds a way to distinguish them is one step closer to there not being any more ways left.

Resisting global traffic analysis for the purposes of deanonymization is not so easy. The issue is that if every time Alice sends ~476MB of traffic, Bob promptly receives ~476MB of traffic, it's not hard to deduce that Alice is talking to Bob. To fix that, the amount and timing of the traffic Bob receives needs to be independent of the traffic Alice is sending him. Which is possible but inherently comes at an efficiency cost.

Well, first thing, you don't send many MB directly to someone. You put it on a Tor onion file-share site, and PM the link via Tor. Unless your adversary is lucky enough to pwn the guard for that onion site, they can't even see the correlation.

I can imagine a global passive adversary that could log all Internet traffic, and make it searchable. The NSA can somewhat do that. But even the NSA can't retain everything for more than a few days, if even that. So even retaining necessary data for a match would be a stretch. Let alone having the processing power needed to do the matching.

> Unless your adversary is lucky enough to pwn the guard for that onion site, they can't even see the correlation.

We're talking about an attacker that can see every byte going over the wire, encrypted or not it's still able to measure the volume of data itself.

> Well, first thing, you don't send many MB directly to someone. You put it on a Tor onion file-share site, and PM the link via Tor.

In that example the attacker wants to determine the location/IP of the hidden service itself. It's pretty well known that high traffic / volume hidden services are some of the easiest targets for global traffic analysis.

> I can imagine a global passive adversary that could log all Internet traffic, and make it searchable. The NSA can somewhat do that.

They don't have to log the data itself simply the meta data and in specific the volume of data sent between vertices in the graph. Various agencies have openly admitted they keep this information and it's not considered "protected" in their view.

This is simply a graph analysis problem, similar to how Bitcoin is pseudo-anonymous unless you use specific methods that aren't built into the core protocol.

Sure, the NSA can arguably see every byte on every wire. But that wouldn't do them much good, unless they knew what to compare with what, over what time period. But OK, say that they had the processing power to compare every traffic stream with every other traffic stream, with time offset up to a week or so. That still wouldn't tell them whether Mirimir had shared something with someone else. Because we'd all be using Tor.

And about onion file-sharing sites. With OnionShare, you can create a site just for that transfer.

What you're describing is equivalent to a dead drop, which is an old school anonymity method. It's not that it doesn't work, it's that it's completely independent of Tor and it doesn't work in all the contexts where Tor is supposed to work.

If Alice uploads the file and Bob downloads it immediately then you haven't gained anything. Bob has to wait some time for it to work, which means you didn't actually need a low-latency anonymity network to begin with. You can't use it for things that actually need real-time communication, like for live streaming or anything interactive.

True. But Tor is the large anonymity network that we have. I2P is cool, but it's too small. For high latency, perhaps Mixmaster remailers via alt.anonymous.messages is still workable. Last I checked, there was a Tor onion news server.

I agree about unworkability for real-time communication. But there, you want to keep messages small. And use padding. Trying to make live streaming anonymous is nontrivial.

I'm not against Tor but I think it's an important fact if we're going to discus how using multiple relays is a way to obfusticate traffic.

What about i2p?

I don't personally know anything about i2p but my spidey senses suggest that until it becomes about as widely-used as Tor, it's not going to have as good an anonymisation factor as Tor.

I mean, if only 100 people use I2P, and some traffic has gone over I2P, you can quite easily narrow it down to the set of 100 people.

I'd be delighted to be proved wrong.

Also problematic is that from my understanding i2p is not used by any honest actors, since it doesn't have clearnet access. Tor is used by lots of people for honest things or at least less-than-completely-shady things like getting around IP blocks or geo-fencing whereas i2p is used exclusively for people that require perfect privacy for their communications.