distributed traffic patterns (for personal traffic)

John Case case at sdf.lonestar.org
Sun Nov 7 21:15:28 PST 2010


Virtual Private Servers are getting fairly cheap, and Amazon EC2 instances 
even cheaper.  A hundred or so dollars per month could put together a 
fairly large set (say, five ?) of general purpose unix nodes on at least 3 
continents ... EC2 in Asia and Ireland, and then some bullshit VPS 
provider(s) here and there in the US.

So let's say you assemble a little quiver of these root shells, and your 
intention is to completely obfuscate your own personal traffic ... to just 
disappear completely (as an atomic, individual net user).

The technical foundation is pretty basic - just set up your own system, or 
perhaps a firewall at your home/office to block all traffic except for SSH 
or HTTPS and to take all outbound traffic and tunnel it over one of those 
to a random one of your systems.  Easy.

-----

So at this point, what do you have ?  First, your own ISP no longer sees 
anything but encrypted traffic.  I suppose they also see time patterns, 
but you've pretty much excluded your own ISP from all information 
regarding your net usage.

Further, you've upped the ante considerably - an actor has to have a 
global eye view of the entire global Internet to collect a true portfolio 
of your net usage.  Maybe a few US nodes are easy, and maybe if you go EC2 
Ireland, that's easy too ... but some VPS provider in Hong Kong ?  Seoul ? 
Sao Paulo ?

-----

So where does this start to fall apart ?

First, if you hit any kind of personal/vanity/small sites on a daily or 
hourly basis, an attacker just has to camp out upstream from there and 
collect all the source IPs that come in.  So if you run your own 
mailserver (or whatever), this falls apart almost immediately.  You either 
need to start hosting a lot more traffic on your own server (giving away 
accounts or something like that) or you need some kind of jumphost that 
isn't connected to you, and an out of band link between the jumphost and 
your own site(s).  This isn't rocket science, though ... just set up a big 
ubuntu/freebsd/whatever mirror next to your own vanity host, and cross 
connect with an ethernet cable over non routable IP space.

Second, unencrypted login sessions ... between web forums and chat rooms 
and any number of other things, somewhere you're entering a user/pass over 
plain old HTTP... and if an attacker can guess one or more sites that they 
know you visit, they can, just like above, camp out upstream and just 
collect all of your proxy IPs.  This is easy to fix with a firewall rule, 
but may be impractical to deal with that way ... a certain discipline with 
nyms and aliases, etc., needs to be adopted, and unlike typical "nym 
leakage" you need to be thinking of nym leakage on an infrastructure 
level.  Not impossible, though...

Where else does this fall apart ?  I don't think there is a weakness with 
DNS resolution, since the dns resolution will come from your proxy, which 
will move around... presumably your local connection has everything but 22 
(or 443) blocked.

Remember, this isn't Tor ... the idea here is not to create mathematically 
provable security.  The idea is to make ones own traffic disappear to a 
degree that one needs to be a nation state to put it back together.  The 
local ISP can't see it, the servers you visit can't see it, and the 
intermediate ISPs (and even a conglomeration of several of them) can't see 
it.

And it looks like this is a fairly cheap thing to put together in 2010...





More information about the cypherpunks-legacy mailing list