[kepler-dev] SDF/PN question

[Stephen Andrew Neuendorffer]

At 11:33 AM 7/26/2004, Bertram Ludaescher wrote:

>Hi:
>
>Efrat and I are just having an interesting discussion on how the
>display actor (and similarly, a new "logging actor") work/should work.
>
>Assume you have a display actor to which two parallel actors are
>connected:
>        ________
>[A1]--|DISPLAY|
>[A2]--|DISPLAY|
>        --------
>
>In an SDF domain, it seems that input channels are tried in the order
>the corresponding actors where connected -- let's say A1 first, then
>A2.
>
>If the token arrives first on A2 (even a week earlier ;-) then nothing
>would be displayed until also A1 arrives (because DISPLAY cannot fire
>until A1 is ready).

Correct.


>Conversely, if we change this to PN, the behaviour would/should change
>(I think), right?

Actually no...  It sounds like what you want to plot is the 
'non-deterministic merge'
of the outputs of A1 and A2, which is not possible to express in PN, or DDF.
Why not just connect them to different displays, or display them against 
other data,
such as 'wall-clock time'.

 > Or would another director (Efrat just mentioned DE) be more appropriate?

Yes, other domains can express a non-deterministic merge, such as DE or 
CSP.  Also,
some variants of dataflow (notably the "PN-like" model of YAPI) allow 
components
to test for the presence or absence of inputs.  It IS arguable that there 
are times
when such semantics are desirable, but we have not implemented them, 
preferring to
stick to deterministic styles of dataflow.

Steve