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

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



At 1:37 PM -0400 9/18/02, Michael Richardson wrote:
>   I am of the opinion that for any system which has a non-trivial number
>of defined connections, that it pretty much impossible for Bob to determine
>that he isn't under attack.

For both the 4/6 and always-4 choices, it doesn't matter if you are 
actually under attack: it only matters if being under attack causes 
you to run out of resources. If you are under attack but have plenty 
of resources, it doesn't matter.

Thus, the only thing that a system needs to determine is whether it 
is getting low on resources. In 4/6, if you get low on resources, you 
go to 6 mode. In always-4, you start re-using your DH exponents.

>   The only deployments that I can see that could determine for certain
>that they aren't under attack would be two or three node VPNs.

Lots of systems could determine if they are getting low on resources, 
particularly if we know that the resources are the ability to do 
large multiplies, and free space in the fragmentation buffers.

>   As such, I question any design that introduces complexity to optomize the
>unusual case.  I would rather have shorter setups times, but I would rather
>it was resistant to DDoS.

There appear to be (at least) two types of complexity: complexity of 
the protocol, and complexity of extending the protocol. For both the 
4/6 and always-4 choices, the protocol complexity is the same ("do I 
have any DH exponents ready for this new connection?"). However, as 
this thread is pointing out, the extension complexity for 4/6 appears 
much less than for always-4, which is why Jan asked why are changing 
IKEv2 to use it.

--Paul Hoffman, Director
--VPN Consortium