3 ms·
What I'd probably do is split the problem -- confidentiality of messages can be ensured locally, and directory services, persistence, and foiling traffic analys
by rdl 13y ago
What I'd probably do is split the problem -- confidentiality of messages can be ensured locally, and directory services, persistence, and foiling traffic analysis on a network service. Network service is long-lived and has much more bandwidth. It can do cool stuff like autorespond, format-convert, etc. for you. If it gets compromised, you lose traffic analysis protection, but not message confidentiality (which is end to end). You could self host the server portion or pick from a bunch of different providers.
Do traffic-analysis-resistant wasteful routing among "servers". Run constant-bandwidth from the server to the client to foil traffic analysis if you really care about that (say, messages delivered only every polling window, and every polling window, a communications of the same size/direction. That seems pretty impractical on mobile, though, so maybe use some kind of push protocol for "you have a new chat request" and then foiling traffic analysis within the chat session (when you're actively using the device). Unfortunately if you can monitor all of the network, an attacker could identify a recipient just by initiating a request and watching for a new chat opening. You could also do things like block off known travel times or other times when someone won't be in contact, correlate with message response times, and at least be able to rule in or out certain suspects (it won't help in fishing expeditions, but if you have 10k people it might be, you could probably identify a person in a few months of passive analysis).
In short, low-latency is the enemy of traffic analysis protection. A true surveillance proof system looks more like anonymous remailers and less like Tor. But, people want realtime.