Re: Terminology and scenario (wasRe: [Monami6] about draft-montavont-mobileip-multihoming-pb-statement-05current)

[marcelo bagnulo braun <[e-mail]>]

El 26/07/2005, a las 16:19, Ryuji Wakikawa escribió:
>>>> I will only consider CoAs, now.
>>>> Suppose that you have a given visited network, which is multihomed 
>>>> to
>>>> ISPA and ISPB, and each one of the ISPs has delegated a prefix i.e.
>>>> PrefA and PrefB
>>>>
>>>> The MN arrives to the visited network and configures one CoA from 
>>>> each
>>>> prefix, let's say PrefA:MN and PrefB:MN
>>>>
>>>> The situation then is:
>>>>
>>>>
>>>>         +---------------------+
>>>>         | Rest of the Internet|---------|
>>>>         +---------------------+         |
>>>>           |                 |          +----+
>>>>        +-----+          +-------+      | CN2|
>>>>        |ISPA |          | ISPB  |--|   +----+
>>>>        +-----+          +-------+  |
>>>>           |                 |     +----+
>>>>         +---------------------+   | CN1| PrefB:CN1
>>>>         | Visited Network     |   +----+
>>>>         |    PrefA    PrefB   |
>>>>         +---------------------+
>>>>                 |
>>>>              +-----+
>>>>              |  MN | CoA1: PrefA:MN
>>>>              +-----+ CoA2: PrefB:MN
>>>>
>>>> Now consider the case where the link between ISPB and the rest of 
>>>> the
>>>> internet fails
>>>> Which addresses does MN has to use in order to maintain a
>>>> communication
>>>> with CN1? and with CN2?
>>>>
>>>> With CN1, the only path available between CN1 and the MN is through
>>>> ISPB, so the MN must use PrefB:MN as CoA for this communication
>>>
>>> I think PrefB:MN can be selected according to the same mechanism of
>>> IPv6 source address selection. (i.e. by Address longest match).
>>>
>>
>> right
>>
>>>> With CN2, the only path available between CN2 and the MN is through
>>>> ISPA, so the MN must use PrefA:MN as CoA for this communication
>>>
>>> It depens on where the HA is located, but..
>>
>> right
>>
>>> When the MN attempts to use PrefB:MN, it fails to send CoTI/HoTI and
>>> will
>>> not receive CoT/HoT.  Therefore, it will give up using PrefB:MN and
>>> will
>>> pick up PrefA:MN. right?
>>>
>>
>> ok, this is interesting, since the HoTI/HoT CoTI/CoT message could in
>> fact be used, as you mention, as a mechanism to explore paths
>>
>>> So is there any possible issue here?
>>>
>>
>> i don't know, but maybe it is worth exploring a bit further...
>> The point in this scenario is that you need to use different CoAs to
>> communicate with different CN.
>
> OK.
>
>> I mean, only the following combination of address work:
>> - PrefB:MN, CN1
>> - PrefA:MN, CN2
>>
>> The other two combinations, i.e. PrefA:MN, CN1 and PrefB:MN, CN2 don't
>> work
>>
>> So, when the MN is in RO mode, it would need to use different CoAs 
>> with
>> different CNs, which i am not sure it is supported...
>> in particular, which CoA will the MN register in the HA in this
>> scenario?
>
> To send CoTI for both PrefA:MN and PrefB:MN, the MN must register the
> CoAs to HA (i.e. MCoA). The MN may stop using RO mode and sends
> packets through its HA.
>
>> I guess that only one CoA will be reachable from the HA, depending
>> whether the HA is in ISPB or somewhere else.
>
> right. However, the MN will know the destination unreachable by
> receiving ICMP error messages.  Then, the MN should use different HA
> (i.e. HoA) for the unreachable destination.
>
>> But in any case, maybe more inteligence in the CoA selection may be
>> needed for this case...
>
> If we need to consdider the case where HA lost reachability to a
> destination,

ok, this is another good point that is that the MN through one of its 
CoAs can reach the CN but the HA cannot.


>  we have to consider a nice HoA selection mechanism. But i
> think it can be achieved by MN receiving ICMP error messages.
>

First of all, imho having a fault tolernace mechanism (such as a 
multihoming mechanism) based on ICMP unreachable mechanism seems not a 
good idea (AFAIK, relying in ICMP has been discarded in MIPv6 design 
and in multi6 design, becuase ISMP are filtered too often to rely on 
them)

So i would not count on ICMPs
I would use them as hints, but i wouldn't asume that they will be there 
and base the mechanism on it.

Second, i guess we need to explore the different scenarios in more 
detail but at least, i would like to know if we agree that the 
reachability property is related to an address pair, an not to a given 
HoA or CoA.

I mean, it is not good to say, "ok, i will use this CoA because it it 
reachable", but what would be more appropriate would be to say "this 
CoA is reachable from this particular CN address" and maybe for a 
different CN, a different CoA is needed.

The idea is to talk about working address pairs, rather than a 
reachable address

Is this good?

regards, marcelo


> ryuji
>
>> Regards, marcelo
>>
>>> regards,
>>> ryuji
>>>
>>>
>>>> (this is site multihoming example adapted to mobile host 
>>>> multihoming,
>>>> sorry for that, but there maybe better examples that express this
>>>> situation in a more relevant fashion for this case)
>>>>
>>>> Regards, marcelo
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>> Nicolas
>>>>>
>>>>> _______________________________________________
>>>>> Monami6 mailing list
>>>>> [e-mail]
>>>>> https://www1.ietf.org/mailman/listinfo/monami6
>>>>>
>>>>
>>>>
>>>> _______________________________________________
>>>> Monami6 mailing list
>>>> [e-mail]
>>>> https://www1.ietf.org/mailman/listinfo/monami6
>>>>
>>>
>>
>


_______________________________________________
Monami6 mailing list
[e-mail]
https://www1.ietf.org/mailman/listinfo/monami6