Re: OR delays

[Roger Dingledine <arma@mit.edu>]

On Mon, Mar 10, 2003 at 10:40:01AM +0000, Andrei Serjantov wrote:
> Roger,
> 
> I am thinking about packet counting and timing attacks and would like to
> the minimum delay an Onion Router indtroduces in processing a packet.
> Any ideas?

Gosh. We haven't really tested it for fast networks. The current code
generally processes about 1100 incoming data cells per second (140
kilobytes, which is what my dsl connection can do) using about 100 ms
of cpu time.

I just tried it out downloading from moria to moria, and it looks like
we process about 23-24 cells per ms. For example, 4699 cells in 199 ms,
4319 cells in 183 ms, and 5259 cells in 223 ms. So that's nearly 700
kilobytes of cells coming in, using less than a quarter of the second
to process them.

The number of milliseconds per second spent processing cells looks
nonlinear then (1100 in 100ms, 4699 in 199ms), so who knows. I'm afraid
it will be hard to determine how long cell processing takes. What did
you have in mind actually?

Note also that we still have unknown (and known) things in our code that
sometimes eat up 10-20ms without needing to.

And of course, remember that by far the majority of the delay will be
introduced in network delays.

--Roger