Ted,

Some clarifications:

Ted Lemon wrote:
> Sent: Tuesday, September 14, 2010 4:07 PM
> To: Tony Cheneau
> Cc: [email protected] Group; [email protected]; Droms; [email protected]
> Subject: Re: [CGA-EXT] [dhcwg] Follow up request for review of draft-
> ietf-csi-dhcpv6-cga-ps
> 
> On Sep 14, 2010, at 6:08 PM, Tony Cheneau wrote:
> > So, as long as the public/private key are generated on the node, I
> > would say that you will not need to communicate the private key (in
> > clear or in a ciphered form).
> 
> Actually, this is clear as mud to me.   What is the big compute job
> that's being done on the server, then?   Can you point me to the RFC I
> should be reading that explains all this?   Sorry to be obtuse.   :'}

If you have a look at the CGA generation algorithm described in RFC 3972 you 
will see that there are potentially two distinct computationally intensive 
tasks:

1. generation of public-private (RSA) key pair
2. generation of a CGA modifier value for a given Security Parameter (Sec) as 
per RFC 3972. (The Sec is used to harden the resistance of brute force attacks 
on the 64-bits IIDs of CGAs.)

#2 above might become computationally intensive for values of the Sec higher 
than zero (if it's zero any Modifier satisfies 3972). Back in 2002 I remember 
running the algorithm on then standard personal computers (e.g., x86 / ~1GhZ) 
and it would took ~20 seconds for Sec = 1 and a couple of minutes for Sec = 2. 
I do not remember going higher...

I understand the proposal is to offload #2 to the server while #1 would not be. 

IMHO it does not any make sense to offload when Sec is set to 0 because the 
computational load is nil. 

When Sec is set to 1 the computational load of #2 seems to be in the same order 
of magnitude as for #1, which does not necessarily means that offloading is 
useless as #1 might not need to be done by the constrained device itself (e.g., 
it's been offload to someone else, and result is stored on a smartcard.)

That being set I have not been convinced yet that there's a real world use case 
for Sec values higher than zero.

HTH,

--julien
_______________________________________________
CGA-EXT mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/cga-ext

Reply via email to