[Tinyos-help] CTP parent switching

[Omprakash Gnawali]

On Thu, Aug 28, 2008 at 1:14 AM, Michiel Konstapel
<m.konstapel at sownet.nl> wrote:
> Following up on my message about the link estimator, I was also wondering
> about CTP's parent switching threshold. As is, CTP will switch parents if
> the new parent offers a path ETX that is 1.5 hops better than its current
> parent, if I am not mistaken. Now, this is set fairly conservatively,
> because a node would not replace a two-hop, 2.4 ETX parent with a direct,
> perfect link to the sink. However, "in the back" of the tree, where lots of
> link ETXs are added up, you don't want a low threshold or nodes would switch
> parents all the time.

I want you to point to some related discussions about this:
http://mail.millennium.berkeley.edu/pipermail/net2-wg/2007/000723.html
http://mail.millennium.berkeley.edu/pipermail/net2-wg/2007/000729.html

> Now, would it be an idea to base this threshold on a percentage of the path
> ETX? For example, at 20%, a node with a path ETX of 2 would replace its
> parent by one that offers a path ETX of 1.6, whereas a node that has a path
> ETX of 10 would only switch if the new path has an ETX of 8 or lower. That
> way, it would favour optimal routes near the sink, where traffic is highest,
> and favour stable routes further down the tree. A cap would probably be
> needed on top of the percentage, so that it doesn't become impossible to
> switch in very large networks.

It is definitely an interesting idea. You will notice that MultihopLQI
uses a similar approach. We haven't tried this for CTP. What did you
observe in your experiments?

- om_p