RE: [MIP-QOS] RE: MIP-QOS MIP QoS Mailing List is Active Now

Hello, =

Here are some = comments on the=20 requirements :

2> I believe = this can be=20 accomplished with micro mobility protocols which hide the = mobility to=20 the home network. About the delay to establish the new QoS path, my = feeling=20 is that it can be done before the IP handoff by some sort of = predictive=20 reservation scheme (which would be relatively easy to set up in a = micro=20 mobility domain). The problem here lies when the mobile moves = between=20 two micro mobility domains. The QOS path may take more time to = be set=20 up but once again some predictive path could be determined in = advance.=20

Using micro = mobility has many=20 advantages one of which is that the qos path is = not changed in=20 the core (which can then easily use Diffserv since the path is not = highly=20 dynamic) when you move within a domain. In the domain itself, it = seems=20 feasible to maintain more states and therefore deal with a = higher=20 handoff rate.
[Morrow, Glenn=20 ]

Most of what you say is = true and I=20 suspect will likely be supported by the end solution. Your comments = are in=20 line with requirement #4. I was thinking of the macro problem = between=20 administrative domains where the signaling goes as far as it needs = to go at=20 the macro level and across administrative domains. I believe that at = least=20 bandwidth used must be accounted for on handoff even in the wireline = case=20 between a small enterprise with 802.11 (perhaps T1 access into = the=20 premises) and a cellular infrastructure that would have the = characteristics=20 you describe. The point is that your comment is pointing to the = solution=20 space of requirement 4 with a large cellular core=20 infrastructure in=20 mind.

4> Micro mobility also = answers the=20 scalibility problem. Maybe the micro mobility domain can be handled = with a=20 Diffserv mechanism but i believe the cells themselves should = maintain more=20 states. This comes to the conclusion that you can use Diffserv on = the wired=20 part of the network (micro mobility + core) but probably not on the = wireless=20 link.
[Morrow, Glenn ] I think we are = saying the=20 same thing with different terminology. I tried to write = the=20 requirement without referencing the specific semantics of = a=20 solution.

6> I don't = understand=20 this. Using the mobile's home address is not fine ?
[Morrow, Glenn=20 ] Hmmm - again think macro when the COA changes and how RSVP = indexes=20 it's state tables and how security indexes SAs and the extent = to which=20 handoff can occur between access technologies at both the micro and = macro=20 levels served via separate autonomous systems of cellular and=20 enterprise.

7> More = protection could=20 be provided if the medium has a a high error rate. In IEEE 802.11, = many=20 transmission rates (with a more or less robust code -- high rate is = less=20 robust of course) are used according to the link's quality. Is this = what is=20 meant in this requirement ? What media should be considered ? =
[Morrow, Glenn=20 ] In cellular, as the voice codes go over the air, = portions of the=20 code words that are "more important" than other portions of the code = word=20 are actually duplicated and FEC'ed (forward error correction = mechanisms=20 applied) before they are transmitted over the air. This duplication = and=20 FECing of these "very important" parts of the code words is commonly = referred to as unequal protection. If unequal protection was not = used, your=20 cellular calls would sound like dog-poop and the cellular operators = would=20 likely not be able to charge as much and an operator who did turn = this on=20 would likely get more customers and more revenues than an operator = who=20 didn't.

One of the excuses = of having to=20 have a layer 5 entity (ex. SIP) intimately associated via signaling = with the=20 routing equipment on the visited network is this need to have the = cellular=20 access equipment perform unequal protection on the voice codes as = they are=20 sent over the air. If you ever want to get rid of this coupling for = cellular=20 networks, then your reservation/binding scheme will have to convey = the codec=20 used and the routing equipment will have to be put into the loop of = this=20 signaling. Another solution people have proposed is to sniff = the=20 packets all the time - this isn't very good. In order to do = unequal=20 protection you must know the format of the code words i.e. the = codec=20 being used. Knowing a traffic class or the delay or the bandwidth = won't=20 work. There are many codecs and sometimes the codec actually = consists of=20 multiple codecs. I hope this answers your=20 question.

When=20 we get into the non-mip contengent of the IETF, this will definitely = come up=20 and a layer 5 solution may end up competing due to this requirement. = I'd=20 rather people get these issues out on the table and be upfront about = this=20 earlly on.

8> Is two way = really an=20 issue ? Maybe for the ACKs ? RSVP deals with duplex links as 2 = simplex=20 links.
[Morrow, Glenn=20 ]

It is an = issue in that=20 sometimes it is desirable to reduce setup delay = and bandwidth=20 used over air for signaling when the optimization can = occur. From=20 my understanding of NSIS, the solution may or may not be based on=20 RSVP. If RSVP can meet the requirement then it can meet the=20 requirement.

9> = Same comment as=20 8>
[Morrow, Glenn ] Agreed. I'll = bet we can=20 fit both into one.

14> I believe = this depends=20 on whether we talk about IPv4 or IPv6. It seems to me that the = solution is=20 not the same in both cases. When talking about = IPV4/IPV6=20 being dealt with seperately, i think the principles should be the = same (esp.=20 if route optimization is a requirement). However, IPv6 has some = specific=20 functionnalities which we should take advantage of.
[Morrow, Glenn=20 ]

There are actually = some=20 security issues with the current MIPv6 spec pertaining to this = functionality=20 as well. I know this is a shocking thought but whatever NSIS = turns out=20 to be may superseed even MIP. This is because NSIS is dealing with = all of=20 the requirements from the get go. I can actually think of a = signaling=20 methods for IPv4 that would allow application packet delivery, the=20 initiation of a binding and a reservation. The question is how = far does=20 NSIS want to go. Also see my question below about same solutions for = both=20 layer 3 technologies.

Gwendal LE = GRAND
[Morrow, Glenn=20 ]

Thanks Gwendal, = these were some=20 excellent comments and insights.

Glenn

-------Gwendal LE GRAND-------
mailto:Gwendal.Le-Grand@lip6.fr<= /A>=20 tel: +33 (0) 1 44 27 75 12
http://www-rp.lip6.fr/~legrand =20 fax: +33 (0) 1 44 27 74 95
Universite Pierre et Marie Curie, = Laboratoire=20 LIP6-CNRS, Bureau C646
8 Rue du Capitaine Scott, 75015 Paris,=20 France