Re: [6lowpan] Call for working group adoption (Re: making progress on fragmentation)

[Carsten Bormann <[e-mail]>]

On Aug 18, 2009, at 15:37, Ralph Droms wrote:

> does anyone have empirical evidence about the impact of fragment  
> loss in 802.15.4 networks that would motivate the need for reliable  
> fragment delivery?

(no WG chair hat:)

I can't answer that question.

But it is important to describe the applicability of http://tools.ietf.org/html/draft-thubert-6lowpan-simple-fragment-recovery-06 
  some more.

As it stands, the draft addresses the mesh-under case.  There is  
little gain for the hop-by-hop fragmentation/reassembly implied by  
route-over; it is also not clear how to apply the draft to the route- 
over fragmentation optimization I described in the other sub-thread  
(who would send the FRACK?).

We do have "fragment recovery" in 4944 in the sense that it is (lower- 
case) recommended to use 802.15.4's link-layer ACK to increase the  
reliability of each (L2) hop (4944, Section 2, first paragraph).

What we don't have is recovery on the whole L2 path from the  
fragmenting L3/L2 boundary (mesh-under ingress, MUI) to the  
reassembling L2/L3 boundary (mesh-under egress, MUE).  The argument I  
have understood for the draft is that there may be (MU) route changes  
(possibly including MU multipath routing) between MUI and MUE during  
transmission of the fragments of a single (larger) packet, causing  
some parts to be lost even with good use of per-hop acknowledgements.

A 1280 byte packet will take some 40 ms of pure airtime on 2.4 GHz or  
some 500 ms on .868 GHz.  Multiply this by a realistic media sharing  
multiplier, and there does appear some opportunity for radio  
characteristics to change during that time.  But that is just a hunch,  
and some numbers from a real-life network would indeed help.

For now, my (WG member) answer to the WG chair's question whether we  
should adopt that draft is a luke-warm "I wouldn't mind".

Gruesse, Carsten