I would like to work towards the adoption of Pascal's
fragmentation draft (draft-thubert-6lowpan-simple-fragment-
recovery) as a working group document. Partly for the
reason giving in the draft, that retries are needed to make
delivery of large packets reliable, but also because retries
are needed to make delivery of *small* packets reliable.
With Ember's earlier, non-IP stacks, we have found that
customers expect IP-like reliability even from non-IP
protocols. We have gotten reliable delivery (as reliable as
UDP packets in the Internet, say) on a radio mesh only by
repairing routes without dropping data packets.
The hard part of route repair is knowing when to do it. The
most reliabile way is to use end-to-end ACKs within the
mesh. Using NACKs would be more efficient that ACKs, but
these are lossy networks and NACKs are even less reliable
than data packets, because they only get sent when something
is already broken.
When a route breaks, we get the following sequence of
events:
1. send a packet
2. fail to receive an ACK
3. update the route
4. resend the packet
This has allowed us to get end-to-end mesh reliability
approaching that expected from IP links. The usual
higher-level IP retry mechanisms can take over from there.
(Sending fragments multiple hops in a route-over network
will require some kind of label switching. That will need
to be discussed over in ROLL.)
How close are we to consensus on Pascal's draft? What
needs to be fixed/improved/removed?
-Richard Kelsey
_______________________________________________
6lowpan mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/6lowpan
