Anonymity on mobile devices

[Christian Beil]

Hi,
developing on a project which uses Tor, I hope to get some opinions from 
you.
I'm working with the mobile business group at my university in Germany. 
We are developing a platform for location-based and context-based 
applications. We also want to provide security and anonymity to the 
users of these locaion- and context-based services. Beside using 
pseudonyms, we want to apply an anonymizing service like Tor.
Our tests with some quite fast mobile devices (PDAs) showed that Tor 
could not (yet) be applied directly on the client. In the first place 
performance of the PDAs is too low for the (many) publice key 
operations, and secondly setting up a circuit causes much traffic which 
takes long and costs money; e.g. the OR list is quite big.

So we switched to a different architecture: now there is gateway to 
which the user connects to and which does all the anonymizing for him. 
This means we have a single point of failure, but we only need to 
connect securely (TLS,VPN,...) to the gateway.
Additionally we want to enable the user to choose the way of 
anonymizing, e.g. using Jap or Tor. Because of this and because we use 
the gateway for some other things, we had to design our own protocol 
which is similar to Socks, but has some additional parameter for the 
anonymity configuration.
So our architecture looks like this: the mobile client connects securely 
(by VPN) to the gateway, then it sends a Socks-like connect request 
along with the configuration parameters to the gateway, the gateway 
sends a request to the chosen anonymity service (e.g. talking socks5 to 
Tor on port 9050) and after the connection has been established the 
gateway forwards all incoming data.

What do you think of this architecture and of anonymity on mobile 
devices in general? There was a system called mCrowds which implemented 
Crowd's Jondos on WAP-gateways. Does anyone know it?

Christian