On 09/16/2016 06:57 AM, Tom Henderson wrote:

On Thu, 15 Sep 2016, Robert Moskowitz wrote:

5206-bis specifies how to user RVS for the 'double-jump' mobility problem.

3.2.3 1) says:

1. The mobile host sending an UPDATE to the peer, and not receiving an ACK, MAY resend the UPDATE to a rendezvous server (RVS) of the peer, if such a server is known.

But it DOES know there is an RVS IF the I1 had FROM and RVS_HMAC parameters and it had created a VIA_RVS parameter to send in the R1.

Yes, but the responder may not know the initiator's RVS even if the the responder's RVS was used, and it also may be the case that neither host's RVS was involved in the session setup.

I see now. As currently speced, R has no way of learning I's RVS. The 'easy' way to fix this is for I to include a VIA_RVS in the I2 packet for mobility support.

"If you every want to get back to me, I can always be reached at this number".

This VIA_RVS provides the knowledge and locator of the peer's RVS.

In fact an aggressive mobility UPDATE would be sent simultaneously to the host and its RVS. If the host had not moved itself, it gets both and drops the one from the RVS.

I believe that Baris Boyvat on the InfraHIP project was looking a while back at such an approach to fast mobility; it was called 'shotgun' approach to mobility and multihoming (try all candidates simultaneously), if I remember correctly.

This comment recommends changes to 5204-bis 4.2.3 that the main goal of VIA_RVS is to facilitate support for the double-jump mobility problem and secondarily "to allow operators ...".

And to 5206-bis section 3.2.3 to use the VIA_RVS to 'know' that there is an RVS for the host and to optionally aggressively send HIP mobility UPDATES to the RVS.

It seems to me that we ought to state that hosts should be prepared to handle duplicate mobility updates sent in parallel to different locators (such as to RVS(es) and to more than one of the host's addresses). We could also state that the aggressiveness of a host replicating its UPDATES to multiple destinations, to try them in parallel instead of serially, is a policy choice outside of the specification. Any other comments on this possible change?

- Tom

Hipsec mailing list

Reply via email to