Re: [Ipsec] Asymmetric Security

At 2:24 PM +0200 2/14/05, Yoav Nir wrote:
>I would add a fourth scenario, which is actually a variation.
>
>(4) Asymmetry in clustered gateways
>Let us say that there are several gateways (call them A1 and A2) 
>that appear to be a single gateway (i.e, have one identifier, one 
>certificate, one IP address) with some load-sharing clustering 
>hardware or software.  Any peer (call it C) must never see anything 
>other than a single gateway (call it A).  The problem is that due to 
>performance constraints, the two gateways cannot synchronize their 
>state after every packet, and because of the retransmission counter 
>of ESP, they need to have two SAs, one for A1 and one for A2.  In 
>IKEv1 this was a problem, because IKE implementations were not 
>required to support multiple redundant SAs (and to C, it looks like 
>the two SAs are redundant).  Such implementations always deleted one 
>of the SAs.  In IKEv2 C is required to support multiple redundant 
>SAs.

As you noted, it is not generally feasible for two devices to appear 
to be one device, due to sequence number management constraints. 
Also, there is this small matter of synchronizing the keys generated 
for the SAs between the two devices, in a secure fashion.

So, a solution today is to create distinct SAs to each device, and 
recognize that they represent parallel SGs and don't try to make them 
look like one SG.

What would you want IPsec to do differently, and what implications 
have you discovered as a result of any proposed changes?

Steve