Switched to 6lowpan from ROLL...
Carsten Bormann a écrit :
[...]
(If you think that routing within a subnet is not something that
immediately comes to mind when hearing "route-over", I'm with you,
but again that is the current terminology.)
TErminology may be so, but I haven't agreed with it, and I believe it
could be subject to change when there's a couple of different queriers.
With the backhaul model, in a mesh under network the link and
subnet are equivalent as the link spans the entire LoWPAN.
But do you mean when there's no backhaul model
(I don't tend to think about the case where there is no Edge Router
-- is that what you mean?)
I meant when there _is_ an ER and the entire "subnet" covers both the
wireless lowpan nodes _and_ the wired ER interfaces. That's what the ND
doc says, that there's a single "subnet" for both. In this case, ER and
6lowpan nodes doing "IP routing" within a single subnet is difficult to
understand to me.
the link and subnet are not equivalent?
In a pure mesh-under world (which the cited text is talking about),
they are equivalent -- making them equivalent is exactly the job of a
full mesh-under protocol. In a route-over world (with or without
embedded mesh-under islands), they aren't.
It's strange...
In this document, a LoWPAN subnet is defined to be a collection
of LoWPAN links interconnected by routers that have the same
subnet prefix.
Let's exemplify collection with two members:
How does a router connect two links, yet has a single interface,
and has the same subnet prefix on both (on both "what")?
("Two links" sounds as if the extent of a "link" is independent from
one's point of view, which in a radio network it isn't. A 6lowpan
node typically has exactly one interface and thus one link.)
MAC layers should offer a link abstraction to the IP stack. I could
have two WiFi APs in a room, all radio hear each other, yet there are
different IP links, allowing to build a typical IP subnet on each such link.
By making the on-link decision (i.e., L2-address the host via ND vs.
L2-address the next-hop router via the routing protocol) based on
the routing information instead of just using the prefix.
A routing protocol based on L2 addresses... is it an IP routing protocol?
Is that a dumb repeater?
No. Forwarding of a packet received is based on L2 address matching
(possibly followed by reassembly) followed by a consultation of the
L3 FIB.
I get the first part "forwarding based on MAC address exact match" but
if forwarding is done so then why consulting the L3 FIB?
Is that an application-layer (UDP) repeating all packets it
receives? (no routing table entry, only one interface).
No. The forwarding is at L3, e.g., no interpretation of IPv6
fragment headers or destination options headers is involved in the
forwarding.
But you said above the forwarding is based on L2 address matching, thus
not at L3... I'm sure there's a logic in what you said, and I may have
missed something by too much dissection.
But it would be very helpful to make a picture about a single-interface
router presented with a packet in input, name the examined field, name
the type of search (exact match?) and name a typical entry in that table.
Alex
_______________________________________________
6lowpan mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/6lowpan
