On Fri, Dec 2, 2011 at 1:59 AM, Lorenzo Colitti <lore...@google.com> wrote:
> On Thu, Dec 1, 2011 at 09:07, Tao Sun <hisun...@gmail.com> wrote:
>
>> You can already do this today with RIOs, and I don't see why there
>>> need to be two ways of doing the same thing. Adding a different method to
>>> configure the same information is twice the code and more than twice the
>>> complexity, since when you have two sources of information with different
>>> semantics you need to figure out how to merge them, what happens if they
>>> disagree, what to do if one is wrong, what to do if one expires, and so on.
>>>
>> We can not do this today. As 1 and 2 mentioned, both the deployed gateway
>> and the 3GPP specification need to be updated to use RIO.
>>
>
> What I meant is that a protocol and IETF perspective, you can do this
> today using RIOs. So the protocol is there if 3GPP wants to use it. The
> 3GPP specs need to be updated regardless of whether we decide to
> standardize the proposed DHCPv6 route option or not.
>
We cannot only discuss the aplicability in IETF perspective. It is indeed
significant change. The operator always needs to consider the huge number
of legacy device and network elements. Change user's device and update
network elements are really non-trival things.
>
>
>> I agree we need to consider the disagree/confliction among policies.
>> That is caused by the multiple interface feature. If you use RA for all the
>> interfaces, the rule received may still conflict. Addressing the
>> confliction is exactly what MIF WG shall consider IMHO.
>>
>
> It is not caused by having multiple interfaces, it is caused by the fact
> that there are potentially two sources of information. It can happen even
> on only one interface, if the network on that interface supports both RAs
> and DHCPv6 route.
>
For one interface in a single administrative domain, the network can pick
one mechanism. Even if two ways are used simutaneously (I do not see this
is necessary though), the administrator shall make the routing policy.
Again, I still think regardless which mechanism is used. MIF shall address
the issue of multiple sources of informaiton.
> Also, in the mobile case I don't understand how you would use a
>>> different gateway anyway. The mobile network is a point-to-point link, so
>>> there's no way to distinguish multiple gateways. How would you use this?
>>>
>>>
>> There are two types of scenarios. 1) A host is connected to the mobile
>> network and WiFi at the same time, a typical scenario of MIF. 2) A mobile
>> host connects to multiple APNs simultaneously, i.e., multiple PDN
>> connection is established. The type 2) scenrio is also considered to be
>> addressed in the OPIIS (Operator Policies for IP Interface Selection)
>> work item in 3GPP.
>>
>
> #1 can't be solved using either the DHCPv6 route option or RIOs because in
> general the carrier network doesn't know the IPv6 address of the default
> gateway on the wifi network.
>
I don't understand why one interface need to know the address of the
default gateway on the other network. If neither DHCPv6 and RIO cannot
solve this MIF scenario, what solution can solve it?
> #2 can be fixed both using DHCPv6 route option and using RIOs.
>
Here you *can* do this using RIO, but limited in IETF protocal perspective.
>
>
>>
>>> ***Broadband network
>>>>
>>>> 1. WiFi network. Some WiFi hotspots provide local services. The
>>>> route configuration on RG is needed to direct some traffic to local network
>>>> while other traffic to the Internet.
>>>>
>>>
>>> I don't understand why you need host changes to do this. Since the
>>> hotspot router is a router, it can simply forward the packets the right way
>>> by itself.
>>>
>>>
>> You may refer to the Section of 3.4 of this draft or Section 5.2 of RFC
>> 4191.
>>
>
> I don't see how Section 3.4 of this draft ("applicability to routers") is
> relevant to this discussion. Section 5.2 of RFC 5191 describes a different
> scenario (two routers on the same link) than the one you suggested here
> (one router on a given link).
>
Section 3.4 related with your previous question, the hotspot router is a
multi-homing router which may need some specific route.
>
>
>> 2. VPN network. When a user connect to enterprise VPN network,
>>>> the routing of VPN traffic need to be configured. Due to the large number
>>>> of such VPN network, we cannot assume all the VPN network only use RA.
>>>> DHCPv6 provides another choice which may be preferred by the VPN network.
>>>>
>>>
>>> The rationale here seems to be that "if we create a new option, some
>>> networks might use it". That's certainly true, but it doesn't provide a
>>> compelling reason why we need the option in the first place.
>>>
>> To us, its a compelling reason due to the network managment/operation,
>> per-subscriber service, foresee burden of modification to existing gateway
>> and specifications.
>>
>
> Who is "us", and what is the use case? I don't see why the ones discussed
> so far can't use RIOs.
>
Us means operators owns and manage both cellular network and WiFi netowrk.
WiFi plays more and more important roles in mobile operator's network.
_______________________________________________
mif mailing list
mif@ietf.org
https://www.ietf.org/mailman/listinfo/mif
