[Tinyos-help] CVS update and Collection problems

[Philip Levis]

On Sep 24, 2007, at 9:34 PM, Iñigo Urteaga wrote:

> Hi,
>
> Thanks for the appreciation. However, I still have the compiling
> errors related to Collection.

I thought it might be useful to make a few notes about the current  
state of collection, because it's really the first time there's been  
some cross-WG collaboration and we're trying to figure out the way to  
handle CVS when there are some experimental extensions that look very  
promising.

The net2 working group proposed adding a new interface to link-layer  
communication, based on experiences developing link estimators for  
collection. This interface provides the "white bit," which allows a  
higher layer to ask if the channel had high quality during a packet.  
Because this would be part of the packet HIL, it's core's  
responsibility. So core and net2 started talking about the tradeoffs.  
Core agreed that it sounded like a good idea, but wanted experimental  
evidence that it helped before making it a HIL.

In a series of experiments, net2 showed that the white bit can  
significantly improve link estimation. For the gory details, you can  
read the technical report SING-07-00, which is a draft of a  
submission to HotNets.[1] The short story is that, in comparison to  
MultihopLQI, the new CTP link estimator reduces routing cost  
(transmissions per packet delivered to base station) by 29% and is  
much more effective in sparse topologies. Based on these results,  
core gave the go-ahead to take the next step, which is to demonstrate  
that the white bit can be implemented on different radio chips and  
helps more than the CC2420.

Incorporating the white bit involves some significant changes to CTP  
(the net2 collection protocol); trying to maintain a parallel CTP was  
seen to be too big a deal, especially given the benefits we've seen  
with CC2420-based platforms. So Om and Rodrigo checked in the new CTP  
version. The recent collection problems are due to the need to iron  
out some of the lingering issues. For example, since the white bit is  
not yet an HIL, it's not part of ActiveMessageC; this means that the  
link estimator needs #define statements to link to the right radio  
chip. If the white bit joins the HIL, then this cruft will go away.

Anyways, that's the situation.

Phil

[1] http://sing.stanford.edu/singpubs.html