[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: [DNA] Considerations for DNA Schemes with multiple Interface andLayer 2 Technologies
Actually the DNA scheme would better not be the same depending on wether i am able to attach to more than one link or not. Effectively, if i am attached for instance to two links at the same time, i may not need to go into the DNA process whenever i receive a hint that one of those links (L2 or L3) is no more valid, if i am able to switch my flows onto the other link without noticeable service interruption (after checking that my IP configuration on that other link is valid).
This is why i do not really like the sentence
"DNA schemes are typically run per interface. When a host has
multiple interfaces, the host separately checks for link changes on
each interface."
As said in the introduction of the dna goals document. I think DNA would better run on an overall basis i.e seeing all of the links (which my Ip layer is capable of for that matter).
Eric Njedjou
-----Message d'origine-----
De : owner-dna@ecselists.eng.monash.edu.au [mailto:owner-dna@ecselists.eng.monash.edu.au] De la part de Pekka Nikander
Envoyé : lundi 25 octobre 2004 08:59
À : Dna
Cc : Erik Nordmark; YGHONG; ???; Greg Daley; JinHyeock Choi; ???
Objet : Re: [DNA] Considerations for DNA Schemes with multiple Interface and Layer 2 Technologies
>> I think that the case 3.2, 'Single interface with Make before Break'
>> will affect DNA solution design. I agree that we are not familiar
>> with this case, so it would be of much help, if you elaborate the
>> below with more detail.
>>
>> ... But in CDMA[7], a
>> mobile station can access two or more different base stations
>> simultaneously. In this case, a host can make a new link-layer
>> connection before an old link-layer connection is torn down.
>
> I hope that L2 can actually tell the packets apart.
> If this is possible, then the implementation could present the packets
> from the different links are arriving on different interfaces to IP.
I agree with Erik and others that the apparently cleanest way is to model the situation with distinct (virtual) interfaces, each corresponding to a wireless connection.
However, I am also wondering how often we need that in practise.
As far as I've understood (which may be completely wrong), in the typical case the CDMA network models the whole base station network as a single link, with a single excess router. OTOH, I have no idea whether soft handovers between operators is defined or not.
If they are, then we probably need some care there.
Taking a slightly wider angle, we can imagine various anomalies like someone inadvertedly connecting two (wireline) LAN segments together. That would (temporarily) create a similar situation.
Hence, I'm afraid that we need to make the DNA scheme(s) robust against these kinds of situations. However, my current opinion (subject to change) is that we should not worry too much about these kinds of situations, and it would be perfectly OK if it takes some more time to system to recover in such a situation. In other words, if some hosts unnecessarily run DNA and autoconfiguration, that wouldn't be too bad.
Network operators have to live with the limits of technology anyway. If our DNA effort results in a piece of technology that has a number of well documented limitations, that is OK, as long as the WG agrees on the limitations.
--Pekka