Re: [netlmm] Consensus call: PMIP6 IPv4 support change

["Laganier, Julien" <[e-mail]>]

Hello Sri,

Sri Gundavelli wrote:
> 
> Hey Julien,
> 
> But, we did discuss the need for UDP encap for control plane even when
> NAT devices are not present in path. Please see the thread.
> 
> http://www.ietf.org/mail-archive/web/netlmm/current/msg05166.html

Please see more below - 

> To repeat, Using  L4 delivery semantics, will provide a proper delivery
> path for control messages to the MIP modules. Carrying both control and
> data packets natively will require in-kernel interceptions and some
> implementation complexity, vs. receiving a packet on transport socket.
> Other reasons, on path firewall devices, non MH semantics in IPv4 headers and
> and the use of transport ports...

Just above you talked about the MIP modules. One of the primary customer for this work is 3GPP. The 3GPP EPS system architecture supports both DSMIPv6 and PMIPv6, and the HA and LMA are located in the same box (PDN-GW). In the IETF, the decision to base Network-Based Mobility on Mobile IPv6 signaling was motivated by the simplicity of implementing both in the same box, as in the 3GPP use case. By making the PMIPv6 and DSMIPv6 encapsulation different you complexify the implementation. 

As to receive the packets in kernel, that is required already when PMIPV6 is transported over native IPv6, so I guess that's doable for native IPv4 as well. 

> I see you are not comfortable with this, I understand. But, we
> discussed this issue. We really need to close this work.

Sri, we did discuss a similar, but different issue, on which we did conclude, but the assumptions were different. The issue was, if NAT traversal is required, let's not re-invent the wheel and reuse DSMIPv6. 

Now for the sake of simplicity we say don't need NAT traversal, there's no practical usecase, so I'm just asking for consistency.

So far I have not seen a convincing explanation of why UDP encapsulation should be used, I am still left with 1) handling firewalls that apparently can't be configured to let IPv6 in IPv4 traffic pass thru, and 2) making life easier for developers who can intercept IPv6 packets in kernel but not IPv4 ones.

Best,

--julien