On Jul 30, 2009, at 9:21 AM, Mikael Abrahamsson wrote:
On Thu, 30 Jul 2009, Fred Baker wrote:
Um, what does a router do? Look at the example in the text and ask
yourself if you want an average user (my canonical "average user"
being my daughter, who wanted me to come to her house to install a
camera on her computer so she could use it on Skype - "did you try
plugging it in?") manually installing routes in each of the four
routers when they could in fact learn them from each other directly?
So, looking at this from another angle, namely deployment. I'm a
router engineer, I support the use of routing protocols as much as
the next router engineer, but I think a good question to ask is
whether most home CPE vendors think RIP for IPv6 is hard to
implement, or if this is something they consider easy?
If it's easy to implement RIP for IPv6 then I'm a proponent for that
model.
Linksys and Cisco (which in this context represent completely separate
product lines, one based on ODM sources and one based on IOS) both
support RIP; the IOS products support all relevant IGPs.
Fred, (just checking) the model you're advocating then is that
DHCPv6-PD from the main home CPE (with WAN connection) hands out
subnets which are then announced to all home gateways via RIP(v6) ?
What I am suggesting is that:
a) DHCPv6-PD is very likely used by the ISP to delegate a prefix
to the CPE (although I know of providers that would prefer to
use an option on an RA to do this)
b) the CPE router installs a default route to its upstream (duh)
c) the CPE assigns some of the implied /64s to its interfaces as
described in 2.1 and 2.2
d.1) in a home with one router, that being the CPE, we're done
d.2) <something> may, in cases like the 2.3 case, sub-delegate
prefixes by some algorithm to other routers in the SOHO/SMB.
e) in a SOHO/SMB with multiple CPEs or with internal routing,
routers will need to communicate within the SOHO/SMB about
routes. For that, I would suggest the use of an IGP routing
protocol as opposed to manual static configuration.
As I observed Tuesday in v6ops, (d.2) is largely undefined; one could
use DHCP-PD, but one could also imagine something akin to or using RFC
2894, or other approaches. Section 2.3 clearly has the allocation
process (which could be in the CPE Router and probably is, but doesn't
have to be) have some notion of a map of the SOHO/SMB; 2.3 is tree
structured and therefore pretty simple (the CPE allocates a /63 to
each of three downstream routers), but a network with a cut-set of two
(which I would hope any self-respecting SMB would deploy) would need
something more comprehensive. For example, in
/-------+-/ /
prefix:2| |
+---+--+ |
|Office| |
|RTR 1 +--+ --
+---+--+ | +-------+ /
prefix:3| | |CPE RTR| |
/-------+-/ +--+ISP 1 +------ ISP 1
| +-------+ |
/-------+-/ |p \
prefix:4| |r --
+---+--+ |e
|Office| |f
|RTR 2 +--+i
+---+--+ |x
prefix:5| |: --
/-------+-/ |0 +-------+ /
| |CPE RTR| |
/-------+-/ +--+ISP 2 +------ ISP 2
prefix:6| | +-------+ |
+---+--+ | \
|Home | | --
|RTR +--+
+---+--+ |
prefix:7| |
/-------+-/ /
Figure 3: Complex SOHO
if prefix:6 and prefix:0 are in fact the same wired network or 802.11
SSID, we would want the "Home Router" to figure that out and use
prefix:0 without trying to allocate ...:6 to it.
--------------------------------------------------------------------
IETF IPv6 working group mailing list
ipv6@ietf.org
Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------