Informatics: Resource-efficient data management

How do you train an existing database system so that it can also process and store large quantities of graph or network data? The kind of data that is growing ever more important in today’s connected world – in the realms of navigation or in social networks, for example – and whose description defines relationships between objects in sober scientific terms as edges and nodes?

For his doctoral thesis, informatics expert Professor Marcus Paradies concerned himself with the storage and processing of precisely this kind of graph data in database systems. The thesis was submitted at his alma mater, the Technical University of Dresden, but was produced in close cooperation with the software company SAP in Walldorf.

Enlarge

“It was mainly about storing and processing network data from the energy sector,” Paradies explains. “The aim was not only to develop software to integrate the data in a main memory database system: Above all, we wanted sufficient performance to allow large data sets to be processed quickly.”

That, he adds, is possible only if you know exactly how the hardware on which you are running these database systems works. Paradies insists that you need to know what the hardware can and cannot do: “The software must be developed in a way that maximizes the utility of the hardware – but also in a way where users do not have to take care of this themselves.”

Research should lead to practical benefits

It is important to Paradies that his research delivers practical benefits. “That is why it is vital to work with real use cases and real data from science and the economy, to validate the practicability of the strategies you develop.”

Regarding scientific data, the professor has high hopes of accessing a wide range of projects in and around Munich – including cross-faculty projects at LMU. “Almost every field of science today works intensively with data,” he says. “Take medicine or bioinformatics, for instance. This is a good place for me to start, because the problems addressed are domain-specific but – from the perspective of informatics – very similar in nature.”

Managing earth observation data

After earning his doctorate, Paradies moved to the German Aerospace Center (DLR) in Jena. Here, he worked primarily with large volumes of earth observation data. “It is very important to understand the earth system,” he explains. “And the ESA’s Copernicus program gives the scientists there a vast pool of high-resolution data. Even if they are only dealing with one small aspect, such as logging activities in the Amazon rainforest, they are already dealing with huge quantities of data that would be completely unmanageable without the support of informatics.”

His team’s expertise is equally vital to the modeling of individual measurement data in ionosphere research, which looks at solar storms that may lie ahead: “We want to give the researchers tools with which they can inspect the models visually and interactively so that they can be compared with each other – in time series, for example.”

Back to research

Paradies ended his time at DLR as head of a department where he was in charge of about 50 staff. “Above all, though, I wanted to do academic work, research, which is difficult due to all the administrative work you have when heading such a large department,” he says. He applied to LMU in the full knowledge that appointment processes take time. So, in the meantime, he lectured for nearly a year at the Technical University (TU) of Ilmenau, his home city.

Even after taking up his current position at LMU, however, he has continued his collaboration with DLR: Together with the DLR Institute of Robotics in Operpfaffenhofen, for example, he conducts research into the management of telemetric data from robots.

“Here, we work with time series data documenting what the overall system status of a robot was before it dropped a cup, for example. As an informatics expert, I don’t necessarily have to understand why the robot made a mistake. What I have to do is provide the tools that enable researchers in this domain to work effectively with the data and both pinpoint and reconstruct such rare mistakes among huge mountains of data.”

Keeping an eye on resource efficiency

Besides managing data on modern hardware, Paradies is interested in the optimal deployment of hardware in terms of economic resource efficiency. “Developing software for the cloud instead of for local servers gives you elasticity and, with that, a billing model that can be flexibly adjusted to the required compute and storage resources,” he explains. “That way, you only pay for the compute and storage resources you really need.”

Yet another aspect of his academic portfolio concerns what is known as green computing. The objective is to engineer software that works not just quickly but also energy-efficiently. “Especially in the AI community, people have so far tended to have the attitude that it costs whatever it costs. But as AI models grow ever larger and their appetite for energy increases, the community is now rethinking this view. Software must be designed for performance, but it also has to take account of rising energy needs and costs.”

Paradies chose LMU because “it is very strong on both applied and basic AI research. I see lots of potential to join this up with interdisciplinary research and transfer projects in the future.” He is convinced that his systemic approach has identified a useful niche within LMU’s impressive informatics setup. “And as I said, there are also plenty of opportunities to work into adjacent fields and join with subject area specialists and AI experts for joint projects.”

Paradies clearly has no doubts: “There is also the fact that Munich has become established as a key European AI hub in recent years. In fact, it ranks as one of the top AI addresses worldwide. However you look at it, this is an attractive venue.”