> I would love to know more about these applications.

Section 3.2 notes the prevalence of compile-time options to reduce the size of 
various 6LoWPAN implementations, which implies that users required this 
functionality for various applications.


> I would also like to discuss code sizes in a quantitative way — how big is a 
> 6282 decompressor?  (You can already leave out most of the compressor..)

While it is true that you can leave out most of the compressor if you want to 
do so, this does not actually result in huge savings, given that implementing 
the decompressor requires many of the same subroutines that the compressor 
requires. In table 3 of the paper of the paper we provide some code size 
measurements, which vary some between stacks. It is a difficult to isolate 
exactly how much code size can be attributed just to 6282, however, as most 
implementations intertwine fragmentation and compression to some extent, and as 
it is hard to estimate how much the code that calls fragmentation/compression 
functions is complicated by the existence of 6282. We discuss this in section 
3.2.

> Maybe it is also worth discussing implementation strategies that don’t follow 
> the (paper-only) model of “compression”, but instead operate on the 6282
> representation right away.

Hmm, I am not sure I understand what you mean by this. Could you explain? Are 
you suggesting a different capability spectrum than the one we lay out?

> Let’s discuss 6282 separately from 4944 and 6775 — the latter just is going 
> through an update that is intended to reduce some of the complexity (while
> introducing some other complications, unfortunately).  I don’t think there is 
> much that can be reasonably shaved off 4944.  Neighbor discovery (6775) is
> a completely different beast, though, and a fresh look at this might bear the 
> most bountiful fruit.

We did find that 4944 is much smaller than the other two, but certain aspects 
of 4944 exist that are not implemented often, such as decompression of the 
broadcast header and the headers required for mesh-under support. Also, 
implementations of 4944 and 6282 are often closely intertwined, which can 
complicate analyzing them separately. I agree that discussing 6775 on its own 
would work well, and might be particularly interesting, though we do not 
analyze it as much in the paper because only two of the stacks we analyzed had 
full implementations.

_______________________________________________
6lo mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/6lo

Reply via email to