[Tinyos Core WG] Meeting: 5/30

Prabal Dutta prabal at cs.berkeley.edu
Thu May 31 13:12:03 PDT 2007


On 5/31/07, Kevin Klues <klueska at gmail.com> wrote:
> > If you assume that you're giving up portability, then instead of
> > parameterizing using val_t, you could provide all of the Read (or
> > Notify) interfaces by the generic component and use a runtime
> > mechanism of specifying which subset to sample.
>
> This is what I was suggesting in my response to Martin's first email.  The
> only disadvantage is that you then have to have all the sensors wired in and
> ready for sampling even if you never plan on using them.  I guess we could
> define compile time constants that #ifdef sections of the configuration to
> include certain sensors and not others if we really wanted though.

Is the main concern here program memory waste from instantiating all
of the generics and associated code for the sensors?

The discussion so far has focused on minimizing the sensor group's
intra-sample jitter but another important consideration is minimizing
the inter-sample jitter (sample-to-sample variation in sample period).
 It seems like the proposed resource model of waiting until all locks
are held, and then taking a sequence of samples, might increase the
inter-sample jitter.

- Prabal


More information about the Tinyos-2.0wg mailing list