Re: [certi-dev] late arriving federate

[Eric Noulard]

2009/9/27 Mathias Fröhlich <address@hidden>:
> Hi,
>
> I try to get up two federates both time constrained and time regulating.
> One starts up immediately enables time constrained gets 0 as federate time in
> the timeConstrainedEnabled callback, past that it calls
> enableTimeRegulation(0, 0.1) gets back a federate time of 0.1 in the
> timeRegulationEnabled callback. Then classes are subscribed and objects
> registered and the regular timeAdvanceRequest/update loop is entered.
>
> At some later time, a second federate joins the federation and executes the
> same code as above with a different federate name and so on.
> The second federate also gets 0 and 0.1 in the appropriate callbacks.

This looks strange to me.

> But then the first federate stops getting timeAdvanceGrant callbacks until the
> second federate's federation time has reached the same numeric value than the
> federation time of the first federate. Then the time advances again in both
> federates.

However, considering the first weird behavior, this behavior looks good.

> I would expect that from within the timeRegulationEnabled call I get the
> actual minimum logical time I need to become time regulating instead of
> stopping the other federates until the new one catches up.

You are pretty right.

> Portico behaves like expected.
> Well in contrast to portico certi behaves very different wrt the federate time
> and lbts and the time values returned in the timeConstraintEnabled and
> timeRegulationEnabled.
> Since I do not have any standard document for HLA/ieee1516 available I am not
> sure how this is supposed to work, but at least I would think that certi's
> behavior is wrong and to me, porticos behavior looks much more plausible.

I'll check the standard regarding this, but intuitively I think your are right.
Pierre, Petr, Martin [Adele],
may be some of you already have a clean answer for this.

> I have done a patch to rtig to return at least something that is about at the
> current logical time of the federation in the timeRegulationEnabled callback.
> But I believe that this change is still wrong since it returns the lbts
> instead of the federate time if I understand right.
> Anyway, I do not really understand the current implementation to its end, so I
> have just attached the change that appears to improve the situation as a
> discussion base for a correct solution.

Whatever the solution, this issue needs to be bug-tracked properly.
Would you be kind enough to open a bug in the tracker for this
https://savannah.nongnu.org/bugs/?group=certi

and attach your patch proposal to the bug.
(you may attach some files to a previously submitted bug
 see the "Attached Files" section of the bug)

> Knowledge, suggestions, hints?
> ... help?

Not much now, just coming back from one "no-network" week :-)
More information to come.

-- 
Erk
Membre de l'April - « promouvoir et défendre le logiciel libre » -
http://www.april.org