RE: Wireless ND was: about violation of standards

["Pascal Thubert (pthubert)" <[e-mail]>]

Hello Phil:

Sure, matching a subnet per P2P link and then route between subnets one way of doing things.  Note that it is a deployment decision, not our decision as std makers or vendors. There are considerations out there like stability of links, capability to configure stuff from a management system and cheer scale that make people go for certain ways or others for their use cases. WiND allows to map subnets on P2P but does not force it, couldn't if we tried. 

There are practical use cases in the wireless world where subnets are not P2P:
-  IEEE std 802.11 infrastructure BSS enables one subnet per AP, and emulates a broadcast domain at L2. Infra ESS extends that and recommends to use an IPv6 ND proxy (per IEEE specs) to coexist with Ethernet connected nodes. WiND incorporates an ND proxy to serve that need and that was missing so far.
-  BlueTooth is hub and spoke at the MAC layer. Should we need to configure a different subnet for each peripheral node (e.g., sensor)? I 'd hope not. Hopefully RFC 7668 allocates a prefix to the central node acting as router (6LR), and each peripheral forms a global address from that same prefix and act as a host (6LN). 
-  Smartgrid networks build route-over subnets of thousands of IPv6 nodes and with such scalability demands, managing a subnet per P2P link is not even considered. Peerings come and go with the dynamics of radio propagation, and only a minimal portion of nodes in range to one another actually peer. The 6TiSCH architecture reflects that model, and generalizes it, e.g., for industrial plants, cars in a parking lot, whatever.

Note: In BLE like in many wireless cases, Link local Addresses need only to be DADed inside the pairs of communicating nodes. In that example, 2 peripherals connected to the same central node can not have the same Link Local Address because the BCEs would collide at the central node and the central node could not talk to both at the same time over the same interface. WiND defines the appropriate DAD operation, but ND does not, because peripherals are no on a same broadcast link. OTOH, Global and ULA are DADed at the Subnet Level.

WiND allows P2P, P2MP hub-and spoke, MAC-level broadcast domain emulation such as mesh-under and Wi-Fi BSS, and route over meshes. There is an intersection where Link and Subnet are congruent and where both ND and WiND could apply. These includes P2P, the MAC emulation of a PHY broadcast domain, and the particular case of always on, fully overlapping physical radio broadcast domain. But even in those cases where both are possible,  WiND is preferable vs. ND because it reduces the need of broadcast, more on that in the intro of https://datatracker.ietf.org/doc/html/draft-ietf-6lo-backbone-router.

Is 6MAN ready for this?

About the size of the registrar, the backbone router enables a distribution of it. But really that is a discussion of the past. Routers can have gigs of memory. The registrar is usually push to a redundant server somewhere, e.g., on modern fabrics that often rely on overlays, mapping servers can scale, to the 10^12 I heard claim for the case of LISP MSMR. The real question is how to migrate from the ND reactive model to a version of ND that is more friendly to new usages such as rapid mobility and large overlays. draft-thubert-6lo-unicast-lookup is a beginning of an answer.

And please do not dismiss the efforts that we had to make to enable routing in that space. RPL is not primitive in my book. WiND and RPL brought in a number of concepts that are now adopted in datacenter routing for RIFT, and though only nascent, I call RIFT a modern routing protocol.

All the best;

Pascal

> -----Original Message-----
> From: [e-mail] <[e-mail]>
> On Behalf Of Philip Homburg
> Sent: dimanche 28 avril 2019 20:18
> To: [e-mail]
> Cc: Pascal Thubert (pthubert) <[e-mail]>
> Subject: Re: Wireless ND was: about violation of standards
> 
> > I started an informational draft for 6MAN on WiND (wireless ND) to
> > explain what physical models it serves and how it can be used over
> > several MAC layer abstractions.
> 
> I read draft-thubert-6man-ipv6-over-wireless-00. I quite like the description of
> the issues with wireless media.
> 
> There is however one thing I miss: The abstract says "This document describes
> how the original IPv6 Neighbor Discovery and Wireless ND (WiND) can be
> applied on various abstractions of wireless media."
> 
> I'm also interested in *why* WiND should be applied. In particular, Figure 2
> shows a complex multi-layered setup.
> 
> So my question is, why not use routing and keep subnets restricted to one
> link?
> 
> I hope there is a clear set of arguments why routing is not sufficient.
> Right now it reads too much like a primitive routing protocol on top of ND.
> 
> Another issue that often shows up with registration based systems (such as
> quite common with IPv4 and wireless) is what happens if a host has more
> addresses than supported by the registration system?
> 
> I.e., it might scale better to have lots of point-to-point links and prefix
> delegation than to have a single big subnet and a hierarchical registration
> system.