I'm not sure that this star-vs-mesh discussion is so important, because
even if you choose the simplest star topology, data propagation is still
required. Configuration of the satellites is simple: *everything* goes
to the hub. But the hub needs to know which satellite to send each
packet to, and as the satellites' configuration changes, the hub
constantly needs to be reconfigured. So it's really the same problem in
a more limited form.
On 03/07/2012 12:54 AM, Yoav Nir wrote:
> Hi Steve
> On Mar 6, 2012, at 11:54 PM, Stephen Hanna wrote:
>> So please review this short document and send comments.
> While the draft does a good job of describing use cases, and certain inadequate solutions, I think it's missing a description of why this is hard.
> Even if we accept the solution of a star topology, where a satellite needs only have one single tunnel, there are really two choices:
> 1. that each satellite know about all the protected networks of all the gateways in the configuration, or
> 2. that satellites send all traffic to the "core" or "hub" gateways. This includes clear traffic (as in HTTP to facebook.com). This increased the load even more.
> If you don't want #2, then the satellite still needs to know about every IP address whether it is protected by some gateway (and therefore needs to go in the tunnel), or not (and so packets with that destination should go out in the clear). Since the protected networks change, this requires that information to propagate throughout the network, and dynamic updates to SPD
> If we don't want a star topology, the gateways or endpoints still need to know what is or is not encrypted. They also need to either know about all peers, or be able to find the peer and (securely) learn how it should authenticate. Either way, without a star topology, you need dynamic updates to PAD.
> I think the draft should mention this.
> IPsec mailing list
> IPsec@... > https://www.ietf.org/mailman/listinfo/ipsec