Zach Shelby a écrit :
Alexandru Petrescu wrote:
Zach Shelby a écrit : [...]
I hear you saying the link-local scope may not cover the
entire link, only the part of it which is fully transitive and
reflexive. It leads me more and more to think a non-transitive
link is actually two links. Each link with its own link-local
scope, each fully transitive and symmetric. (If so, the
problem left is to fit a router with a single interface
connecting to two links simultaneously (the parts of a
non-transitive link).)
You can't split this non-transitive link into a clean set of
transitive ones just like that. It doesn't work, especially from
the single-interface router point of view.
A - B - C
A can reach B, B can reach C, but A can't reach C.
B is a LoWPAN Router. From its perspective, its link-local scope
includes both A and C. It does not have two links by any means.
It is forwarding between two nodes on the same link (and in the
same link-local scope) who don't have transitivity between each
other (A and C).
B in LL scopes of both A and C, B has a single interface... could
it be that B forwards a packet from A to C without A hearing it
too?
A's link-layer can hear when B forwards the packet (unless it went to
sleep immediately)..
Ok.
Node A receiving its own packet could be easily qualified as noise.
However, as the IP destination is the unicast address of C, the
link-layer destination is the MAC address of C. Thus A's link-layer
will discard it.
Ok when the IP and MAC addresses are not multicast.
When they are, like when A sends an RS to "all-nodes" IP address
(converted into "all-nodes" MAC address) B repeats back this packet to A
(because in the same link-local scope as A), in addition to forwarding
it to C. This could lead to more noise, maybe even an RA generated by A.
Medium access control schemes of wireless link-layers are able to
deal with extra transmissions from multihop forwarding.
I agree MAC schemes could alleviate the B transmits back to A problems.
And these schemes are part (supposedly?) of the Mesh-Under concept.
Which makes think that Route-Over wouldn't work without Mesh-Under at
the same time.
Of course because you are forwarding with a single interface, your
channel capacity decreases with more hops (this is the multihop
penalty of all such schemes, including Mesh Under).
Ok.
Otherwise nodes should have a means to know on _which_ link-local
scopes are they, and send data on only one scope. But I think
there's only one link-local address per interface, and only a
scope_id field in the respective C struct.
Not needed, the link-layer deals with this using MAC addressing and
the MAC algorithm. A node just tracks who are its neighbors.
Ok. Some questions rest about how does the link layer deal with
link-layer multicast addresses, when they are used by RS and NS, such
that to avoid noise amplification on non-transitive links.
Alex
_______________________________________________
6lowpan mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/6lowpan
