gye

Terminology is the single biggest problem here: "Anonymous communication" in CS does NOT mean you do not reveal your identity to the person you are communicating with. It means you stay anonymous from the network perspective: No third party learns with whom you are communicating, and everything you reveal to your communication partner becomes an explicit attribute: Your "identity"/name/nickname, but also your location.

Again, very simplified. The design makes sure everyone knows about the same network. The network is being monitored by independent folks constantly, and you cannot just ramp up lots of capacity without anyone noticing. Not only the current state, all past states of the network are recorded.

You also make it sound like there would be an alternative design, like a closed set of operators. Now, that would obviously be worse than an open system. All you need is two non collaborating entities in your path.

Note how I am trying to convince you with facts, whereas you point to my preference of anonymity as something that makes them less worthwhile. Such ad hominem attacks should be avoided in serious discussions, thanks. I value my anonymity.

Where does Tor claim to do more than it can? Can you point at any specific statement?

Compared to anything else in that space, I always liked how Tor does not promise magic, but is actually pretty open about its weaknesses. Most of what comes up as "new discovery of a weakness" is actually already covered in the original design paper and FAQ.

Tor already "pads" to cell size. There is ongoing work towards better padding. Look at WTF-PAD for more. It is surprisingly difficult to improve padding effectively, and you just can't add delays to TCP streams. Tor is not "refusing to add" either of those things.

A more recent and thorough real world analysis of the traffic correlation problem is https://www.freehaven.net/anonbib/#ccs2013-usersrouted . In general, that site has a lot of great papers on these topics.

This can be refuted pretty easily, on various grounds.

One, you are assuming that all parties interested in deanonymizing Tor collaborate. As you say: agencies, not agency.

The design makes sure that you do not need to trust any node in particular. The damage individual nodes can do are small and basically amount to temporary denial of service for non-exits (which clients will quickly route around) and messing with your traffic as exit (which becomes less and less of a problem as everything moves to e2e, and https is improved/hardened with key pinning etc). The remaining problem is statistical correlation, where studies show it is actually not trivial to perform and takes time and repetitive behavior.

Your argument also assumes that everyone is immediately fucked when one powerful adversary can break the anonymity. This will very rarely be the case in real scenarios, where "the NSA" will just not care about you, but plenty of other actors might. By spreading not necessarily false but very simplified claims, you are drawing users who don't know better to systems where they will be clearly much worse off!

The list of relays is public. There are actually surprisingly few exits that carry most of the traffic, which yes, is a weakness, but that weakness is still stronger than anything else you could compare it to. You can look at the large operators like universities and nonprofits like www.torservers.net, their relays, and you will see that it does not take that big of a budget to run the majority of all exits in terms of total capacity. Is is much easier even to contribute high bandwidth entry nodes -- again, feel free to check the largest non-exits in terms of capacity, and investigate the motives. It's all public information.