While computations on averages can prove that a design will not work,
they obviously cannot prove that it will. I submit that the more
interesting data is how long trains of minimum-size packets are, and
how far behind an actual crypto engine can get before it runs out of buffers.
But in a certain sense real data is irrelevant. Tradition seems to
be that one can only advertise "wire-speed" capability if one can
handle a continuous stream of minimum-sized packets, without loss.
Or, at least, only then can one be immune from competitors and reviewers.
Barney Wolff <[EMAIL PROTECTED]>
