From: Adam Back <aba@dcs.ex.ac.uk>
Date: Fri, 2 May 1997 18:22:13 +0800
To: hal@rain.org
Subject: Re: A new system for anonymity on the web
In-Reply-To: <199704261927.MAA02125@crypt.hfinney.com>
Message-ID: <199705020958.KAA00897@server.test.net>
MIME-Version: 1.0
Content-Type: text/plain



Hal Finney <hal@rain.org> writes:
> Adam Back, <aba@dcs.ex.ac.uk>, writes:
> > The disadvantage of crowds is that it won't work to well over a 28.8k
> > modem.  Browsing the web is slow enough as it is, without having
> > traffic make multiple hops through modem users machines.
> 
> Good point.  More generally, a path will only be as fast as its slowest
> link.  I suggested to the authors that the crowds could be segregated
> by access speed, so that there could be separate crowd pools for 28.8
> modem users, for people who share connectivity over a limited-speed line,
> and for people who have their own direct high-speed link.
> 
> Theoretically, at least for long downloads, passing through multiple
> slow links shouldn't slow down the transmission, 

I'm not sure that this is so.

Ascii pic. consider Alice and Bob, without crowds:

        +----------+
        | Internet |
        +----------+
             | 2Mbits
          +-----+
          | ISP |
          +-----+
     28.8k/     \28.8k
         /       \
      +---+     +---+
      | A |     | B |
      +---+     +---+

When Alice and Bob are online, consider that they are actually using
their bandwidth most of the time.  I know I do; if I'm not using it I
hang up, with pay per second phone lines, you're likely to.

Now add in crowds where traffic from A is going through B, and traffic
from B is going through A:

                           +----------+
                           | Internet |
                           +----------+
                                 | 2Mbits
                              +-----+
                              | ISP |
                              +-----+
               14.4k for Alice/     \14.4k for Bob
                  14k for Bob/       \+14.4k for Alice
                          +---+     +---+
                          | A |     | B |
                          +---+     +---+

Now Bob and Alice each have 1/2 the bandwidth they had before.  Add
more users and longer hops and it gets worse.  Of course my claim that
Alice uses 100% of her bandwidth is probably an exaggeration.  There
will be idle times while she is reading.  However there will be times
when both Alice and Bob are downloading, for multiple users with
multiple hops it is likley that there will be more than one user
actively downloading at any time.

> just increase the latency.  I don't have a very clear picture about
> how long it would take to snake in and out of a bunch of people's
> modem lines en route to the web server.

I'm sure it's going to increase the latency too.  My ping times are
200ms from the PPP link alone (ie pinging the PPP server machine
itself).  Add to this that the members of the jondo / crowds pool may
not be on the same ISP, and you've got the additional overheads of
whatever latency is added by the cumulative latency between each of
the hops in the chain.

> If the latency does take an intolerable hit, maybe the idea we discussed
> for the "keep alive" transmission would work, where all the web page
> data gets downloaded with a single request.  (I should clarify that I
> was not suggesting that this be done all the time by default, but rather
> that it be an extension to the http protocol, a different GET request
> or a different parameter sent along with the GET.)



> Another idea, which loses more anonymity but is still as good as the
> Anonymizer, is to run your jondo on your ISP, if that is allowed.  Then
> other people's paths don't go through your modem, they just get turned
> around at the ISP level.  A spy at your ISP may be able to find out
> where you are browsing, but the average web site won't.

This compromise is probably retains a fair amount of anonymity.  The
ISP probably isn't loggin IP traffic, or incoming or outgoing web
proxy (jondo/crowds) traffic, and probably isn't actively attacking
your jondo process to see which bits are for you and which are being
forwarded.

I wonder if you could improve the security of this by trading off
against some additional bandwidth consumption for the ISP.  Say have
split the jondo in half, with two cooperating half-Jondos acting as a
single virtual jondo in such a way that someone with root access on
one machine but not the other, can't extract any useful information by
spying on that half of the jondo.

> > The other problem is the scalability of the strong systems like DC
> > nets or PipeNet.  Bandwidth is still scarce.
> 
> The Crowds people argued that their system automatically scaled, since
> you had more jondos ("web remailers") the more people who used it.  If
> average path length is constant then the anonymous web browsing method
> has a greater cost than the regular way, but it is a constant factor.

Crowds migh scale in this way, but DC nets and Pipenets don't.  Crowds
is a weaker cousin of DC nets and Pipenets.

Adam
-- 
Have *you* exported RSA today? --> http://www.dcs.ex.ac.uk/~aba/rsa/

print pack"C*",split/\D+/,`echo "16iII*o\U@{$/=$z;[(pop,pop,unpack"H*",<>
)]}\EsMsKsN0[lN*1lK[d2%Sa2/d0<X+d*lMLa^*lN%0]dsXx++lMlN/dsM0<J]dsJxp"|dc`