flow label
Fundamentally, we have to decide whether we want to do flow-based
compression in 6lowpan or not.
There is significant previous work in the IETF on flow-based header
compression, but the flows were different (mostly RTP streams in RFC
3095/5225 and TCP connections in RFC 4996) from what I'd expect in a
6lowpan, with significant complexity in the header compression scheme.
Note that much of the complexity comes from maintaining per-flow
context state, assuring that that state is updated in sync between
compressor and decompressor, and robustly dealing with failures in
that synchronization.
The whole idea of CBHC was to avoid the complexity of per-flow state.
Instead we went for a mostly constant, per-lowpan context state, with
an independent dissemination protocol in mind (which we now have as
part of 6lowpan-nd).
On one hand, it is a good idea to isolate the interface between
6lowpan-hc and that dissemination protocol. On the other hand, I'm
not so sure about creating interfaces in 6lowpan-hc to context state
maintenance protocols we don't know (and whose properties we don't
understand). Using the flow label as an index into the context state
(which, from 6lowpan-hc's point of view, is specified as an interface
only) sounds like there is some per-flow state maintenance going on.
To me it would make much more sense to integrate such a flow-based
maintenance more closely with (a then quite different form of) 6lowpan-
hc, as this is less likely to be mostly constant as in the existing
CBHC approach. But maybe I'm misunderstanding how a flow label would
be used.
By the way, we don't have very good ways to compress flow labels in
6lowpan-hc. As these are relatively large and constant in a flow,
they would be prime candidates for compression in a flow-based scheme.
If they are instead to be used as per-flow context identifiers, Pascal
is probably right that the size of 20 bits is somewhat unmotivated.
Gruesse, Carsten
_______________________________________________
6lowpan mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/6lowpan
