[kepler-dev] PN and SDF -- getting the best of both worlds

[Bertram Ludaescher]

Hi:

This is actually a question to the Ptolemy/MoC experts (and I might
have asked sth similar before):

I'm looking for a Ptolemy domain/director/MoC that combines some of
the benefits of PN and SDF.

Specifically, SDF can (and does -- that's our problem) a schedule in a 
single thread. For example, if we have:

  /---B--- ..
A+ 
  \---C--- ..


then in SDF, B and C will execute serially and not in parallel.

In PN they will execute in parallel but PN, being so flexible, doesn't
allow us to do much static analysis (e.g. deadlock freeness!?) and
also might have "execution semantics problems", e.g., how can we nest
PNs?

What seems to be useful is to have an MT-SDF director (multi-threaded) 
SDF that is as restrictive as SDF for modeling and design, but
can execute a dataflow process network in a multi-threaded/concurrent
fashion.

Does this make sens? What's out there in the Ptolemy realm? 


thanks!

Bertram

PS for an interesting application think for example of a recursive SDF 
"workflow" for quicksort. Since SDF allows nesting, it should be
possible to do a recursive block diagram in SDF that implements
quicksort quite directly. The two recursive calls to
"quicksort(small_sublist)" and "quicksort(big_sublist)" could be done
in parallel for example. (Maybe the combination of parallelism and
recursion is a bit too much for starters... but then..why not ;-)