[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Latency Costs of Anonymity


From: Wei Dai <[email protected]>
> [My idea for ATM mixes]
> The problem here is that you'll have to do a RSA operation on EACH 
> packet.  Pretty hard on the CPU...

Yes, good point.  It might be possible to use a stream model where the
separate packets which make up a stream use the same conventional key.
This allows the various packets which make up a stream to be identified
as such by outsiders, but still if there are a large number of virtual
streams going through the network at one time it should be possible to
confuse the streams pretty well.  ("I've got a crazy idea.  Let's cross
the streams!" -- Ghostbusters).  Then you only need to do the RSA work
at setup time, and you need a fast streaming cypher during the
conversation.  This is how the streaming-packet encryption models like
IPSP or Netscape's SSL seem to work.

> > I think it may be more useful rather than speaking of "true" anonymity
> > to think of factor-of-N anonymity.  This reflects the bandwidth costs.  I
> > would guess that, if you have a packet-based video converencing system,
> > that today you could probably get factor-of-2 anonymity with custom
> > hardware, and perhaps even more than that.
> I'm not exactly sure what you mean by "factor-of-N".  I only used "true" 
> to distiguish it from "trivial" anonymity (such as using a pay phone).  
> Of course, anonymity, like security, can only be relative.

By "factor-of-N" I meant anonymity where you can only pin the source of
a message down to one of N possibilities.  It appears to me that many
of the costs will be a function of N.  It will be relatively easier to
cloak your source as one of say 50 possibilities than to make it any of
one in a million.  This is why I suggested that factor-of-2 anonymity
would be the easiest.  The DC-Net concept would allow two users to
share a cryptographically strong pseudo-random stream, and each of them
to XOR their video output with the random stream; then these modified
outputs from each of them are themselves XOR'd together to produce the
joint output.  As long as only one sends at a time, the resulting
stream is their output, but it is impossible for an outsider to
determine which one is sending.  The hardware requirements seem quite
modest and perhaps would be adequate today even for video.

> [My points about limitations on suitability of anonymity]
> This is all very true.  I guess I'm just lamenting the loss of my ealier,
> more naive dream that one day everyone will be anonymous (read
> pseudonymous), and that physical and digital identities will be totally
> seperate. 

I don't think we would really expect everyone to be anonymous all of the
time.  In our personal lives, with friends and family, it doesn't seem
appropriate to expect anonymity (although my earlier quotes from Greg
Bear's sci fi story suggest differently).  But still I think that for
people who desire it and are willing to pay the prices, anonymity would
indeed be available in many or most electronic communications.  So if
that is your desire you should be able to achieve it.


Version: 2.6