3 ms·
So, TOR works in this way, but some other hypothetical ideal system would never work this way. An ideal system would operate in such a way that reading an ordi
by operandroid 8y ago
So, TOR works in this way, but some other hypothetical ideal system would never work this way.
An ideal system would operate in such a way that reading an ordinary public resource requires no single peer. If I want to simply read a wikipedia page, no single individual should be capable of knowing that I requested that page, and be tasked with the entire responsibility of proxying my request. No single peer should even know that a wikipedia page is what I asked for, who I asked for content from, or where I've been and what I've been doing.
Right now HTTPS offers content privacy, but no meta data privacy. TOR offers a degree of meta data privacy, but no guarantee of content privacy. Even with the two combined, the TOR exit node can still understand that someone logged into Gmail, and someone probably tried to send a message with an overall length of ~2MB.
With a strategic vantage point, enough facts could become evident, that TOR would fail to offer enough protection, and still out someone trying to accomplish a secret task with TOR.
It should be possible to provide both content privacy and meta data privacy, so that my solid gold widget factory plans are mailed safely to my vacation home, and no one knows that I mailed them to myself, that I mailed anything at all, or what I mailed if that's what I was doing.
I don't care about how "it" works, if TOR can only suck one way, and can't not suck some other way. I want something that doesn't suck. I already said I don't want TOR. TOR doesn't need to be something else. It just needs to stop telling people it's something better than it actually is.
- eklitzke 8y agoYou can't magically make bridging to insecure services work in a secure way. If you want to use Tor to bridge to unencrypted HTTP, at some point someone needs to actually make an unencrypted HTTP request. And likewise for HTTPS requests. HTTP is a stateless protocol, which is why what you're describing is not possible. You could design a protocol that let you split up requests in the manner you describe, but that protocol would not be HTTP.
- operandroid 8y agoOne can obfuscate an extremely live distributed, obfuscated cache that mirrors public resources, and that's not magic. Just because you have trouble imagining the implementation of such a thing doesn't mean I do, and doesn't make it impossible.