On 06/ 2/10 05:07 AM, Ralph Droms wrote:
Erik - assume, for discussion, the MTU within the RPL network is uniform
(1500 bytes), and the RPL network includes only non-storing routers, so
all the forwarding headers are inserted by the RPL edge router. Could
that edge router manage the ICMP messages to the source correctly; e.g.,
if the edge router needs 64 bytes for routing headers, it sends the ICMP
message with "please limit packets to 1436 bytes"?
It is actually a bit worse, since RFC 4944 says that the 6lowpan
interface MTU is 1280. Thus any router inserting things, or
encapsulating things, would have to fragment the IPv6 packets. In the
case of inserting RH4 without encapsulating with an extra IPv6 header,
this means "proxy fragmentation" i.e., a router pretending to fragment
on behalf of the original host.
I don't know if we want to be that "creative".
In the case of a uniform ROLL network with a fixed MTU sufficiently
large where all the devices at the "edge" of that network know the
special rules, then I think it would work to have the edge uniformly
insert RH4 in packets going "in", remove RH4 in packets going "out", and
redoing the ICMP errors going "out". Redoing the ICMP errors is
presumably similar to what is described in the tunneling RFCs - not only
the mtu field but also the "packet in error" should be redone to remove
all traces of RH4.
Erik
--------------------------------------------------------------------
IETF IPv6 working group mailing list
ipv6@ietf.org
Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------
