On Mar 2, 2007, at 8:15 AM, Ian Chakeres wrote:
Two quick comments inline.
On 3/2/07, 김용운 <[EMAIL PROTECTED]> wrote:
>I don't think LQI by itself should be used. LQI must be matched with
>other indicators to make good decisions.
I agree. But, in my opinion,
LQI is likely to be the primary key.
The reason is saving TCO.
LQI is good for making single hop decisions, but much more complicated
when we talk about multiple hops (routing).
I'd caution against just assuming that LQI will solve all single-hop
problems.
First, LQI is not clearly defined. Radios such as the CC2420 give
multiple hardware indicators of link quality. The "LQI" field that
the CC2420 provides, for example, is not an 802.15.4 LQI value; the
data sheet recommends that you use other values and factors to
determine link quality. I have yet to hear of a good (produces stable
and accurate values) algorithm to compute an 802.15.4 LQI purely from
hardware indicators.
For example, if you look at what current sensor net stacks do, they
essentially scale the LQI such that only very high SNR links are used
(think >10dBm above threshold), leading to routes that are stable but
are far from the shortest path.
Additionally, if you have analog interference sources, then you must
compute LQI bidirectionally. Otherwise, you may be able to receive
packets from a node but it cannot receive packets from you. This is
also a reason why protocols such as AODV run into big problems in
practice, unless you constrain communication to be over a
bidirectional mesh that you've established underneath IP. But that's
the hard part and the actual protocol problem to solve: once you have
it, AODV and its ilk are fairly trivial.
Finally, applying LQI to routing decisions is difficult, given its
fairly vague specification in 802.15.4. One can imagine computing
minimum cost paths using LQI, but it would be good to see what those
calculations are like in practice, and how different 802.15.4 stacks,
with their different LQI calculations, lead to different routing
performance based on the upper layers assumptions on its meaning.
All together, what this means is that mesh routing is almost
certainly needs to have network-level information in terms of link
quality, which it can use to make network-level routing decisions. It
can use LQI as part of these decisions, but it is important to leave
the flexibility to incorporate other information as well.
Phil
_______________________________________________
6lowpan mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/6lowpan
