random generator

[Eric Hughes <hughes@soda.berkeley.edu>]

>I'd like to point out that higher data rates are very very desirable.

And higher data rates are more expensive.  If you're making one time
pads, you need high bit rates, but otherwise you don't.  10Kbps is
overcapacity for personal use.

Let's do an estimate.  Suppose all you use your random numbers for is
to create session keys for socket connections.  Now lets say you need
to open a socket about once a minute.  Since you need, say, 500 bits
for a DH key exchange, that's a bit rate of about 10 bits per second.

One can cache bits coming in from the random number generator in a
ring buffer.  You can make this ring buffer arbitarily large, or even
virtualize it to disk and make it appear as infinite in size.  Then
you could run your generator continuously and always have enough bits
available for use.

If you're using a generator for making session keys, then you just
don't need that high a bandwidth.  

Now the $/bps for the Newbridge chip is much lower, but for personal
use you throw away too many of its bits to make it worthwhile.  This
higher bandwidth chip would be useful in a server of some sort, where
you are making session keys more than once per second.

Proposal: Make this random number generator operate at 100 bps.  If
higher bit rates are the same price, fine.  But a specific design
criterion of 100 bps should be a practical and economical goal.

Eric