[Ns-developers] [ns3] Statistical framework (draft)

[Joseph Kopena]

On Fri, May 16, 2008 at 6:28 PM, Joseph Kopena <tjkopena at cs.drexel.edu> wrote:
> probably not a first spiral goal, but something to keep in mind.  One
> key design decision here is that I lean toward having simulation
> control outside of the simulation scripts, i.e. we boot & run new sim
> programs for each trial.  It seems cleaner & simpler from a sim
> writer's perspective.

Just reiterating that that's a key design decision that there might be
differing opinions on.  One note is that most of the ns2 packages seem
to work that way.

Also, re. discussion of polling variables.  I don't think the stat
framework should be in general polling variables for their value.  The
trace system provides a nice mechanism to be notified as events are
tossed.  However, note that you could write such functionality very
simply in ns-3, and I can see cases where you might want to do that.
We could also provide a calculator object that provided a convenience
wrapper for that.

Simulation termination conditions seem like they should be implemented
within the calculator objects.  For example, the CounterStat object
could have an optional (settable) threshold value which if hit
triggers the simulation to end.  The code Vincent is working on could
also be wrapped into a StatCalculator type object (roughly
corresponding to an observer in ns2measer and ns2/anakoria2), and
provide similar functionality for when the input hits a given CI.

Thx

-- 
- joe kopena
right here and now