[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[16NG] Re: some thoughts on IPv6-over-IPv6CS

Jari, it is indeed a needed feature to have statements "XXX is performed
as per RFC NNNN".  There is value in that and it establishes a basis for
communication.  That is needed.

The Charter puts the document on Standards Track.  I hope that direction
 is taken with an eye on the fact that not all IETF's IPv6-over-foo
documents are Standards Track (optionally see rfc4392 and rfc3572).

IMHO an IPv6-over-foo document to be on Standards Track would be good to
have at least one _new_ recommendation, something like "IID is derived
from a non-48bit id, in this way".  Other example would be conversion
from multicast address to multicast CID, or conversion of IPv6 Traffic
Class to a 802.16 Service Class, or something similar.  Or an
IPv6-oriented network-entry procedure (not a link-layer network entry
procedure).  Or maybe a IANA action.

But please take this as just some thoughts on the document, not
necessarily as a disagreement.  Again the document looks good, it is
needed and I can live with it ok.  It is possible that later work can
enhance the basic features in this document.

Alex

Jari Arkko wrote: 
Alex,

Let me start by saying that I actually did share a little bit of your
unease given that the document has a lot of focus on the configuration and network architecture issues. However, I believe its
a feature, not a bug when most of the spec contains statements like "XXX is performed as per RFC NNNN." There is indeed very little new to implement here, because you run IPv6 as it has already been defined, and you run L2 as it has already been defined. That's great!
Particularly when a lot of different changes and other ways of running IPv6 over this link have been suggested in the past. Given this background, I think it is also important for the specification to talk about the configuration/architecture issues, because those were the determining factor when we decided that we do not actually need all these new enhancements to run IPv6. 

A few examples:

rfc4392 IP over InfiniBand, an Informational RFC describes some
architecture.

rfc3572, IPv6 over MAPOS, Informational, does describe new things to be
implemented, when deriving the IID from the HDLC address.

RFC3831 IPv6 over Fibre Channel, Stds Track, has at least one new thing
that needs to be implemented in an IPv6 stack: derive IID from Nx_Port_Name.

rfc3146m IPv6 over IEEE1394, StdsTrack, proposes at least new IANA
allocations and SLLAO/TLLAO encodings.

Alex




Jari

Alexandru Petrescu kirjoitti:
Raj, Jari, thanks for having this exchange public on the mailing list.

I will precede this mail with a high-level unease I have about advancement of this draft towards a Standards Track RFC. This can be solved by several ways, eg change its name into 'Goals for ...' or maybe change status, or other way one may think of. The draft is definitely well written, complete in its own and self-contained,
a piece of text worth the effort to submit to IESG.



Usually a a Standards Track RFC gives clear directions about what should be implemented. This document doesn't: there's no new message format, no new message exchange, no new issues with other implementation, 'MUST' occurs only two times and it's for lifetimes
in RA - no new lifetime in RA suggested, 'SHOULD' appears once to say MTU should be what 2460 recommends (which is good), 'MAY' never
appears. It doesn't offer a implementer anything special that
should be done about running an existing IPv6 stack over a new kind
of link (802.16 IPv6CS in this case, allow me the term IPv6CS).

Section 2 Introduction describes what the IEEE standard describes. It also describes what WiMax thinks the architecture should be.

Section 3 Terminology describes a new term 'ASN' which is a WiMax term.

Sections 4 and 5 are architectural descriptions, from IEEE documents.

One very important part in Section 4, giving indication to which specific part to use when putting IPv6 packets on 802.16 MAC (ETHCS
or IPv6CS?) is left implementation-dependent (who implements that?
how?) - is there an RA-overIPv6CS enhancement that says use ETHCS vs use IPv6CS.

Section 6 'IPv6 link' makes recommendation about how to configure things, but nothing to implement.

Section 6.2 'IPv6 link establishment' which is the 802.16 network-entry procedure. What should the IPv6 stack implementer write?

Section 6.3 'MTU' says MTU should be what it should be. This is a good goal, but what to implementer means?

Section 7 "IPv6 Prefix Assignment" recommends each mobile to be on a different prefix. This is (1) different than any other IPv6-over-foo documents (and there are other point-to-point links out there not recommending this) and (2) a configuration management
issue, nothing to be implemented.

Section 8 "Router Discovery" is standard behaviour nothing new, nothing to implement.

Section 9 "IPv6 addressing for hosts" - nothing new. I could think
of SeND issues with the MS having the same link-local address on all its 802.16 connections (subnets) but that's all.

These are the reasons I have that uneases. That said, let me address the point where I was cited, see below.

Basavaraj Patil wrote: 
Hi Jari,

Responses inline:


On 1/23/07 1:04 PM, "ext Jari Arkko" <jari.arkko at piuha.net> wrote:

Basavaraj, all

I reviewed the new version of this spec, and I'm generally quite happy with it. I found three small remaining issues which
are listed below:
+-----+ CID1 +-----+ +-----------+ | MS1 |----------/| BS1 |----------| AR |-----[Internet

<http://tools.ietf.org/html/draft-ietf-16ng-ipv6-over-ipv6cs-06#ref-Internet>>>







]
+-----+ / +-----+ +-----------+ . / ____________ . CIDn / ()__________() +-----+ / L2
Tunnel | MSn |-----/ +-----+


Figure 5: The IPv6 AR is separate from the BS, which acts as a bridge

The part about the BS acting as a bridge seems surprising. Is this really the case? A standard bridge function?
I did'nt mean bridge from the "standard bridge function" definition. I will delete it from the document. Basically the BS is an L1/L2 entity, but I do not believe I will not to elaborate on that.

This section presents a model for the last mile link, i.e. the link to which MSs attach themselves.
I would remove this sentence. You need to explain how the link looks like from an IP perspective. And I think you are doing that. Whether there is a distributed or integrated BS/AR at the
other end is really not a key issue. 
Okay. Will work on this text as suggested.

IEEE 802.16 also defines a secondary management connection that can be used for host configuration. However support for
secondary management connections is not mandatory. A transport connection has the advantage of it being used for host configuration as well as for user data.
Are you specifying something about the use of the management connections? If not, take it out.

Not really specifying anything w.r.t the management connection. This came up during discussion with Alex Petrescu and I added it just for the sake of completeness. Within the scope of this I-D, the management connection has no relevance. I can take it out.
Yes, the issue is that 802.16 recommends the RS/RA to happen on a Secondary Management Connection (instead of on a Transport Connection). Clarifications on the list suggested that probably nobody uses a SMC. But that doesn't mean that the IEEE spec isn't saying so.

So one could write in the draft that "contrary to the IEEE Spec, this draft recommends the use of a normal transport connection for an RS/RA". I would suggest that actually. But there may be other oppinions as well.

Another thing that was mentioned in the list, and it was agreed by me and Tim, is that there may be an issue with the indication of how to auto-configure an address: statelessly or statefully. This indicator comes from an RA in an IP-only world but in 802.16 it comes from a link-layer message too. This is an obvious risk for interoperability: what if the BS says stateless but AR says stateful. In this line of thought some may remark (from my experience in other WGs) that people should be capable to configure
their networks ok, and these clarifications shouldn't be specified. But here we talk things to be done at MAC layer vs at Network layer, and maybe it's not the same people doing it.

So I would suggest to at least clarify that IPv6-over-IPv6CS recommends to ignore that link-layer indicator. But there may be other oppinions on this topic too.

Finally, even with all these clarifying remarks I'm making in the inner body of the mail, I think there is still nothing new to be implemented - which is good in a good sense, less work :-)

Alex


Each MS belongs to a different link. No two MSs belong to the same link. 
Duplication. Remove the first sentence.

Okay. Editorial fix.

Will submit a revised version with these changes.

-Raj

Jari


_______________________________________________ 16NG mailing list
 16NG at ietf.org https://www1.ietf.org/mailman/listinfo/16ng







_______________________________________________
16NG mailing list
16NG at ietf.org
https://www1.ietf.org/mailman/listinfo/16ng