"Basically, it's almost always about solving problems faster than before," says CISPA faculty Dr. Sebastian Brandt, who joined CISPA from ETH Zurich in October 2021. By problems, the 36-year-old means the tasks that a computer must perform in order to provide us with data and services at the push of a button, such as the fastest route to the next major city or online shopping at the click of a mouse. Many of these tasks are not processed by individual computers but in small and large networks.
Since the 1990s, distributed systems have become increasingly relevant and the IT infrastructure more interconnected. There are several reasons for using distributed systems. Since several processes can be executed simultaneously in distributed systems, they work much faster than individual computers. In addition, the performance of a network can be easily increased by adding more computers. Some of the problems we are trying to solve today are also so huge, Brandt says, that the amount of data can no longer be stored on a single computer and therefore must be distributed. "That's the case, for example, with some complex problems from different disciplines. But for some of these problems, we would currently need hundreds of years to solve them, even with huge computer clusters."
What makes it so time-consuming is usually not the computing operation itself, but the communication overhead within distributed systems. In networks, however, computers communicate only directly with their respective neighbors. If distant computers want to exchange information, this requires several rounds of communication, and all of these communication steps take time. Brandt analyzes and designs algorithms that need as few communication rounds as possible to solve problems efficiently. "One of my main goals is to understand: What can be computed if you can only access a small, local portion of the input data?"
Knowing where performance optimization reaches its limits is elementary. "In the past, I've often shown lower bounds on how fast something can be calculated. After all, many researchers try to make systems even faster. But if they know that, under the constraints of the model they are working with, there is no algorithm that is faster than a certain runtime, they don't put unnecessary man-hours into a hopeless endeavor," Brandt explains. In his research field, he says, little was known about what was impossible until a few years ago. "However, we have learned quite a bit over the past three to six years. Now we're in the exciting situation where, for some decades-old problems, we finally know exactly how quickly we can solve them."
Before coming to CISPA as a senior scientist, Sebastian Brandt was a postdoc in the Discrete and Distributed Algorithms Group at ETH Zurich, where he also received his PhD in 2018. "I am excited to see what new research questions will arise through the collaboration with my new colleagues. I already have a few ideas, too."
translated by Oliver Schedler
