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

Good morning everybody. I would like to welcome our today's seminar speaker,

Professor Tuan Basten from Eindhoven University. Let me shortly introduce him to you.

He is Professor of Computational Models in the Electrical Engineering Department

of Eindhoven University of Technology, TUE, and chairing the Electronic Systems ES Group.

He is also a senior research fellow at some institute that sounds very familiar to us

called Embedded Systems Institute. We also have here in Erlangen.

Before he started in the Electrical Engineering Department in 1999, he was working at the

Computer Science Department in the Formal Methods and Information Science Groups at TUE.

In 2000-2001, he spent some time as a visiting researcher at Philips Research Eindhoven Information

Processing Architectures Group, but he has also spent research work at the University of Waterloo

and at Carnegie Mellon University in Pittsburgh. His current research interests are in the area of

embedded and cyber-physical systems with a focus on network and multiprocessor systems,

trade-off analysis, design space exploration and computational models. And here, with that, I would like

to welcome Twan Basten here very much and also thanking him for taking over also the task of being

an examiner today in the afternoon of Joachim Falk's PhD thesis. Welcome.

Thank you, Jürgen. Thank you for the nice introduction. Welcome, everybody. Welcome also to people

that might be linking in through video. It's a nice opportunity here to talk here about the work

that we actually mostly do at the University of Eindhoven, but Jürgen already mentioned my link with

the Embedded Systems Institute and actually that combination of basic research done in academia

and transfer of work towards industry is something I think is an interest that we share and that will

also reflect in the talk today, although this is more a bit on the foundations of data flow and

then particularly on coping with the dynamics that we see nowadays in data flow.

A goal, I guess, that many of us share is to make computing in one way or the other predictable.

So the engineering disciplines, we typically try to design a system, reason about a system.

If we build a bridge, we normally know that it holds. For computing systems, we are typically not that

far yet that we can fully predict not only the functional aspects, but particularly also all the

extra functional aspects, timing, energy, all these kinds of things. So the ultimate goal is to make

computing, let's say, as predictable as some of the other engineering disciplines. And then data,

data processing is a very important aspect nowadays. It's increasing in importance.

If you look here, this is just an overview of some systems which are dominated in the region of

Eindhoven, I must admit, but also some Germans. So Germany is strong in automotive. Cars are getting

data processing engines. They have cameras, radar, all kinds of systems on board to assist in

automated driving or semi-automatic driving. So there's a lot of data processing in there.

But in the medical imaging domain, for example, interventional X-ray, you are in real time

combining normal video, X-ray, with all kinds of other data. Printing, or say printers, they actually

have printers that go up to more than 2,000 pages per minute. So it's extremely high speed printing

at high volumes with a lot of data processing going on. So there are many core platforms inside

those printers to process all the data. Gaming, smartphones, radar, ASML, wafer scanner, so the

machines that make all our chips, they measure, so they have thousands of control tasks, thousands of

sensors on tens of processors doing measurements in real time, processing all that data and

activating those scanners. Also there, the acceleration that this machine actually has to

perform is up to 16G, which is four times as fast as a jet fighter pilot, and it has to be accurate

at the nanometer scale. So all data processing. And the challenge that we see is that the processing

is highly dependent on what you are actually encountering. So this is a simple movie that is being

decoded and you see the workload that is being processed here. The workload heavily depends on the

amount of motion that you see in the movie. So if there are high mobility scenes, you see peaks,

if there is no mobility, then you see the workload drop. Traditionally, to make this predictable is

actually to make the design conservative, cope with the worst case. But you see already here that

there are large peaks, so if you are designing a system for the worst case, you get a very conservative