The following content has been provided by the University of Erlangen-Nürnberg.

Thanks for the invitation here to invite Erlangen for this colloquium talk.

The title of my talk is Hardware and System Software Requirements for Multicore Deployment in Hard Real-Time Systems.

So at first I will tell you something about hard real-time demands and timing predictability, timing predictability and COTS processors.

And then I'll go into the achievements of two European community projects.

One was called Merasa and the successor project which is now running is Parmerasa.

And then perhaps you are going a little bit more into system software or whatever.

So the basic idea of this project is that we find increasing demand for functionality in current and in future real-time embedded systems,

which means for us airplanes, cars, but also all kind of machinery like here this big construction machinery of power machine.

And often there's a demand of mixed criticality applications where applications are of high security and safety level, sorry, and high hard real-time level,

while other applications are of less constraints.

And so altogether we need an increase of processor performance as demanded.

And this I hope it works. Let me see because I didn't check it.

This shows what hard real-time means. A deadline should never be missed. Does it run?

No. Okay. Then forget it.

What you can see here is it's a film and there are people running crossing the railways, the railway tracks,

and then a train is just going to run just immediately before they just jump off the tracks.

So it's really hard real-time.

This is another example for real-time. Time is relevant. It stems from the high peak roadmap.

You can see there's a bear, there's a salmon, and the salmon is just missed. What kind of real-time is that?

Yes, that's it. At first everybody looks at it and says it's hard real-time. It's soft real-time because the bear,

if the bear misses the salmon, there will be another salmon. So it doesn't matter. It doesn't really matter.

But it is hard real-time for the salmon.

Okay. So safety-related hard real-time embedded systems require, as we can see from the salmon,

that a deadline must never be missed. The salmon must be quick enough.

And we need a proof of timing requirements by a worst-case execution time analysis.

Or that's what the industrial people of Palmarasa told me because they don't always do a WCET analysis.

That is common for Avionics but not for automotive or something else.

Or at least demonstrate, depending on the criticality of the system, that the implementation meets its timing requirements.

Whatever that means. It could mean measurement-based WCET. It could mean extensive testing and then giving some surplus,

additional time frame on it. So it means not everywhere the WCET analysis is applied, but we think that it should be applied.

Okay. Rainer Wilhelm was here, I think, a couple of weeks ago.

And I'm sure he presented this very famous slide which he developed.

So what you can see is the number of the distribution of execution times over time.

So typically here is the average execution time.

And if you measure, if you just measure, then you get the maximum observed execution time.

But there may be some execution times that could arise but were not in the measurement interval.

All the possible execution times go from here to here.

And here is the worst-case execution time.

But unfortunately in most programs the worst-case execution time cannot really be computed.

So what we can do is to find upper timing bounds and try to get as close to the unknown worst-case execution time as close as possible.

And that is very important. The upper timing bound can be very far away from the worst-case execution time by overestimating, by pessimism overestimating in the tool.

Okay. How to guarantee such hard real-time demands?

There's a static WCET analysis which models the processor and the memory system totally with everything that should be in the real processor.

Also the design faults and so on.

Modeling and potential paths of all the programs must be done.

And then WCET bound is computed with mathematical tools.

One of these tools is Otava. It's a free tool from University of Toulouse.

And just try to remember UPS means Université Paul-Sapatier.

Okay. But it is the University of Toulouse.