1. Link Model Analysis for 802.16 based Networks
IETF 67 – San Diego, CA
05-10 November 2006

2. Goal
To analyze the IPv6 link models that can be applied for the based networks
And recommend preferably one link model for the standardization
(but we don’t want separate IPv4 vs IPv6 models for the same IP link)
Non-goal: make any statements about choice of CS’s.

3. Link Models Considered
NBMA Link Model (which some call the “IPv6 Shared Prefix Link Model”)
Point-to-point Link Model
Ethernet like Link Model

4. NBMA Link Model L3 AR
Shared Link L2 BS
MAC Conn.
Tunnel L1
The AR, Multiple BSs attached to the AR and all MSs attached to these BSs form an IP Link
One or more prefixes are assigned to the link
doesn’t work between MSs
Which means many apps/protocols won’t work without changes

5. Point-to-point Link Model
P2P Link 2
MS1 AR
Tunnel
MAC Conn.
MS3
MS2
P2P Link 1
P2P Link 3 BS
Each connection between MS and AR forms a separate IP Link
Separate prefix per IP Link means each MS sees a different on-link subnet prefix
Everything L3+ works normally within the point-to-point link

6. Ethernet-Like Link Model
MS1
AR1
MS3
MS2
BS2
BS1
AR2
MS4
Bridge
Similar to Ethernet
Emulates multicast with Bridging

7. Dormancy Dormancy is orthogonal to the link model
The BS knows the dormancy state of the MS
The BS can determine when it is appropriate to wake the MS
The design team concluded that this is not a factor that influences the choice of model.

8. Limiting multicast/broadcast traffic
There is a desire to limit “unnecessary” multicast (and IPv4 broadcast) traffic, even when not dormant
Any multicast traffic applications request is desired (not “unnecessary”)
Broadcast and all-hosts-multicast may have unnecessary traffic
For specific protocols (like ARP), it is possible to add intelligence to the BS and/or the bridge to do intelligent filtering
All other multicast traffic already requires that the MS send explicit IGMP/MLD requests to indicate that it is necessary
This includes for IPv6 ND
IGMP snooping is already well known (RFC 4541) and widely implemented in switches
Summary:
IGMP/MLD is periodic, vs DAD is one-time
IGMP/MLD works for all apps/protocols, DAD only works for one
Recommendation: WG should investigate use of MLD snooping rather than other DAD optimizations for IPv6

9. Short Summary of IGMP/MLD Snooping
Host
Switch
Router
Host
IGMP/MLD: hosts explicitly request groups other than the all-hosts-group
Switch intercepts IGMP/MLD packets and creates group/port state
Switch plays an active role upon topology changes (initiate query on topology change)
Sends all multicast traffic on router port
Only sends requested traffic on host ports

10. Documentation/implementation status
Already commonly implemented for IPv4 by late 2000
There were interop problems at that time, since the behavior was not specified.
IDMR/MAGMA WG began documenting requirements and considerations for both v4/v6
Interop problems were addressed
Finally published as RFC 4541
Summary: mature spec exists, even more mature implementations exist

11. Model Recommendations
Ethernet CS:
Ethernet-like model AR
IPv6 CS:
Point-to-point link model
AR needs some method for Unique Prefix Assignment

12. Current Status WG Document
WGLC is over, but not received many comments
Please review and comment before the document is sent to IESG

13. Questions, Comments, Suggestions