Re: [certi-dev] CERTI and Real-Time simulation

[Eric Noulard]

2008/5/27 HADJ AMOR HASSEN <address@hidden>:
> Hi,
>
> I am using 2 real-time simulators (control federate and a graphical
> federate).  I used CERTI to communicate between both simulators. It works
> fine! Also, i synchronize both simulators to start simulating at the same
> time using annouceSynchronizationPoint. It works fine also. For the first
> simulator, i developed a wrapper to communicate with CERTI. For the second
> simulator, i integrated HLA services in the code because this simulator is
> based in C++ language.
>
> This is an overview of my thesis work.
>
> I have some questions and i hope if someone can answer me.
>
> - The both simulators are real-time. Does this mean that we don't need to
> use Time management services? (non regulating or constrained simulator). I
> done a first test and it seems working fine.

What are the criteria for "working fine"?

> The both simulators are non
> regulator and non constrained and so they didn't request time advancing.
>
> - I done another test using one simulator as regulator and the graphic
> simulator as constrained. I used a time stepped simulation. The simulation
> slows hugely.

Can you gives us figures?
(and tell us how you did get teh measures)

How fast is your "not regulator + not constrained" compared to
"regulator  + constrained"?

Can you give us the number of TICK_REQUEST  in both cases?

Which kind of tick call do you use?

tick() --> classical non blocking tick
tick2() --> CERTI specific blocking somehow like tick(0,MAX)
tick(min,max) --> RTI will spent between min and max time in tick.

> Any idea?

You should tell us more details about your realtime requirement?
Do you need a 200Hz cyclic simulation ?
Do you need a sporadic simulation with contrained latency for UAV?
Do you need to preserve causality or is best-effort with eventual loss OK?

> Finally, I think that HLA shows an apparent weakness when it comes to
> real-time simulation applications.

You may be right (for your testcase) but
I think you didn't gives us enough informations to conclude.

We did successfully experiment 10 to more than 100Hz cyclic and
realtime simulation
with CERTI (2 up to 4 federate), it's doable but it doesn't come
"out-of-the-box" from
HLA.

I think high-performance application barely comes out-of-the box, you need
to be careful. You may use an RTOS but if you application breaks
ReaTime (RT) usual constraints
(minimize I/O, beware system calls, examine algorithm complexity, try
to use lock-free data structure...)
you won't get RT behavior from the best RTOS.

>
> Any suggestion, any idea or remark is welcome.

Gives us more details :=)

-- 
Erk