Re: Regarding pre-round trip for stateless cookie (Jan's issue)

[Radia Perlman - Boston Center for Networking <Radia.Perlman@sun.com>]

>>	From: Michael Richardson <mcr@sandelman.ottawa.on.ca>
>>I would rather have shorter setups times, but I would rather
>>it was resistant to DDoS.

I don't think the stateless cookie is very useful for DDoS. It's
only useful for the kind of DOS in which the attacker system is
using fake IP addresses. In a DDoS situation, the innocent zombies
will be using their own addresses.

As for shorter setup times...ironically, the 4/6 protocol might have
lower latency!

a) if you don't care about Bob being stateless until receipt of the
   stateless cookie, then the 4/6 protocol
   would be lower latency, because it will be 4 messages,
   and the messages will have fewer bits to transmit (message
   3 doesn't have to repeat all of message 1 and part of message 2)
b) if you do care about Bob being stateless, then the 4/6 protocol
   would have 6 messages, but unintuitively might still be
   lower total latency because of two issues:
      . fewer total bits to transmit (small issue)
      . Alice and Bob can't be doing their exponentiation in parallel.
        If Bob is stateless, he can't do any computation until receipt
        of message 3 from Alice. The time to do the exponentiations
        of g^b and g^ab might be longer than the transmission time
        of the first 2 messages.
        
Again, I'm not claiming this is a very important issue either way,
and I got the impression from Charlie's message that he's not thrilled
about being pushed back and forth on an issue that doesn't matter
very much. But it *is* interesting that latency can be less with
a 6 message protocol than a 4 message protocol. (at least *I* think
it's interesting. :-)  )

Radia