From: IN%"tcmay@got.net" 6-MAY-1996 01:41:12.77

-- that a model is lacking (I don't mean we don't have some ideas of what's important, but that we haven't filled in the details, figured out what sorts of correlations an analyst can make by looking at packet sizes, sending times, delivery times, etc.)

Hmm... I wonder if genetic algorithms would be a good way to analyze traffic. Admittedly, this may be a matter of my having a (potential) hammer and seeing things to bash with it.

-- the real world situation with remailers is that message volume is probably way too low for comfort (my presentation on remailers at the first CP meeting outlined a need for about 10 mixes, each mixing at least 10 messages of the same size before remailing...and 20 mixes each mixing 30 or more messages is much better...we are most likely far, far below this, for nearly all remailed messages. Fortunately, most remailed messages are either not critical or are being done for novelty, harassment, flaming, etc.).

It should be possible, given a model, to add onto existing remailers a routine that automatically forwards random messages through a random string of remailers and back to itself (or into a /dev/null address; I can see advantages and disadvantages in the resulting loss of information) at a random frequency whose bounds are dependant upon current traffic levels at that remailer. This might be supplemented by information based on statistics gotten from other remailers. I would suggest Raph Levien (sp?)'s remailer list as a basis for the random string of remailers.

(PipeNet-type schemes may help, depending on a bunch of details. So would "local mixes in cabinets," meaning, Web anonymizers with high bandwidth that do their mixing locally. They have to be "trusted," to some extent, but would help a lot. There are some gotchas.)

There is unfortunately a balance between what an operator will be able to take - in regards to bandwidth - and what is needed for a web anonymizer or remailer to work. For remailers, this requirement is greatly decreased by the lack of immediacy needed. -Allen