Bump–bump, bump–bump, bubumpbump. A monitor shows a heart beating irregularly. A technological innovation opens up new possibilities in the analysis of cardiac arrhythmia: the real time MRI. Magnetic resonance imaging is a medical imaging procedure. Now, thanks to a development by Göttingen physician Jens Frahm, moving images from inside the body are possible in real time. Frahm’s procedure, which won the European Inventor Award, uses special technology to achieve an imaging speed of up to 50 images per second: Only a portion of these 50 images are based on measurements; reconstructive algorithms compute the differences between the measured images and fill in the gaps between them.
From this type of hightech application to a mass transit app for smartphones: real time applications are all around us. “Wherever computers interact with the real world today, ‘real time’ is typically behind it,” stressed Björn Brandenburg of the Max Planck Institute for Software Systems. The head of the Real time Systems Group calls this a strong realworld impact. However, real time is defined differently in these various fields, and not keeping up with it has consequences of differing severity. Real time is commonly separated into hard real time and soft real time: with the former, exceeding the time limit has significant consequences; with the latter, only the quality of a service is reduced. In this distinction, Brandenburg finds at least the absoluteness of the assertion dubious: “There is harder real time and softer real time,” said the software specialist. Whether the mass transit app shows that the trolley is five minutes late after a delay of 30 second or 60 seconds is hardly relevant. With the real time MRI, it is more a matter of the fluidity of the movement and not whether it is transmitted on a one minute delay.
However, if Amazon reports that a delay of 100 milliseconds when loading the homepage costs one percentage point in revenue, then we are talking about a significant economic loss. And if an airbag in an accident or an airplane during an automated landing procedure responds to commands after a delay, the consequences are a matter of life and death. According to Brandenburg, the lines between hard and soft are blurred. “I always need to see what restrictions I can tolerate and what steps I may need to be ready to take to remove those restrictions.”
Real time systems are always subject to inherent latency. “This delay is an additive phenomenon,” emphasized Brandenburg. Take the antilock braking system of a car: The sensors check traction, then transmit this information to the wheels, the actuators and, ultimately, to the car’s electronic control unit, which may initiate countermeasures. All of these processes produce latencies that aggregate into an endtoend delay. To minimize this, every step needs to be analyzed and optimized. Large processing capacity, such as that of quantum computers, is certainly helpful. Even edge computing, where certain data is processed at the edges of the system and does not need to be sent first to the core processor, can improve performance.
No less important is the path the data must travel. The distance and number of waystations affect the transmission speed. Experts agree: Industry 4.0, automated driving systems and the Internet of Things, in which large volumes need to be sent over long distances, can only be implemented comprehensively with the powerful 5G mobile communications standard. Data will soon be sent along one particularly long path by EarthNow: The American startup founded in 2017 has announced plans for satellitebased real time earth observation. A comprehensive network of hightech satellites and a powerful transmission system are designed to ensure that users can view events in real time, such as illegal fishing or emerging environmental disasters. Aerospace giant Airbus, which also wants to manufacture the satellites, and Bill Gates have invested in the company. However, the company has yet to announce when this service will be available.
At the Max Planck Institute for Software Systems in Kaisers lautern, Brandenburg is studying the worstcase behavior of real time systems. Is large processing capacity a guarantee of reliable real time behavior? “Yes and no,” said Brandenburg. Large capacity certainly offers good performance on average, but it is not immune to downward shifts. “If a system has good averages, I often have a lot of very good and a few very bad figures,” explained Brandenburg. These bad figures, however, cannot be allowed to occur when releasing an airbag. This is why Brandenburg is conducting what he calls worstcase execution time analyses. What are the worst possible figures from a system when executing a command? To find this out, he looks for the longest path at the programming level and tests it. “Generating a lot of load, finding bottlenecks and applying stress. It’s a rather laborious process.”
All this serves to optimize more critical real time systems such as airbags or aircraft control systems where delays have grave consequences. Because, as Brandenburg laconically puts it, “The real world doesn’t wait for the computer to finish.”