Re: [certi-dev] NextEventRequestAvailable and LBTS - Possible deadlock?

[Eric Noulard]

2010/7/6 Pierre Siron <address@hidden>:
> Bonjour,
> Hello,
> I suppose that the lookahead values are 0.
> I can reproduce this bad behavior with the interactive federate
> of the HLATestSuite.
> Sorry, that works with two TARA (the RTIA can send a NULL message,
> export RTIA_TM=D to see it), but that does not work with two NERA
> (no NULL message sent).
>
> The solution is to implement a GVT algorithm ?
>
> As previously discussed in the TARA and NERA task (#6898),

https://savannah.nongnu.org/task/index.php?6898
This task was closed, I may re-open it or perhaps it would be better
to open a dedicated bug entry for this issue.

The tracker may have missed a link to the reference given by Pierre:
"Zero lookahead and repeatability in the High Level Architecture"
http://www.cc.gatech.edu/computing/pads/PAPERS/Zero_Lookahead_and_Repeatability.pdf

Other refs:
http://www.informs-sim.org/wsc97papers/0373.pdf
http://www.informs-sim.org/wsc07papers/070.pdf

> to do this (GVT) computation could be a task dedicated to
> the RTIG process (a centralized GVT computation, not a
> distributed algorithm, initiated every Delta t).
>
> Until now, the RTIG process just forwards the NULL messages
> to the RTIA of the constrained federates.
> My idea :
> - when a federate calls NERA, the RTIA sends a NULL' message,
> - the RTIG does not forward a NULL' message but it computes
> a GVT,
> - if this GVT grows in presence of NULL', it sends standard NULL messages.


What is the GVT you are refering to?

>
> This could also solve the time creep problem.
>
> Is this algorithm correct ?
> I have a doubt with complicated loops.
>
> Best regards,
> bien cordialement,
> Pierre
-- 
Erk
Membre de l'April - « promouvoir et défendre le logiciel libre » -
http://www.april.org