Factories for data International research on "self-awareness" of data-centric IT platforms
The research project “Information Processing Factory”, which the Institute of Computer and Network Engineering at Technische Universität Braunschweig is involved in, is starting its second project phase. The aim is to master the growing complexity of chips and networked systems in the age of digitalisation by applying new principles. To this end, a consortium of scientists from TU Munich, the University of California in Irvine, Kookmin University in South Korea and TU Braunschweig has taken inspiration from the operation of intelligent factories, in which many related issues can be found.
The concept of the “Information Processing Factory” (IPF) was developed in the first phase of the project. This was applied to individual chips for safety-critical applications, such as those used in autonomous vehicles or modern medical electronics. The focus was on the operation of these highly compact IT systems, where the calculation, transport and storage of data is organised, energy consumption and temperature are regulated, and system status, ageing and reliability must be constantly monitored.
IPF 2.0 in the truck-vehicle-network
In the second project phase, this control concept is now being extended to so-called networked systems consisting of many complex chips. A vehicle network of autonomously driving trucks serves as an application example. Each vehicle is equipped with many networked chips. Together, they assess their environment via wireless communication, recognise and avoid dangers, react to obstacles and coordinate their routes. The application is provided by a new project partner, Kookmin University from South Korea, which has also built a practical setup with model vehicles on the roof of a university building in Seoul.
IPF 2.0 uses the work from the first phase for a paradigm shift in the design of IT platforms, focusing on the robust and networked operation of a large number of complex chips. Such a refocus is not only aimed at the growing hardware problems at the end of “Moore’s Law”, according to which the complexity of integrated circuits regularly doubles. It is necessary in order to master ever-increasing system complexity and data volumes. The interest is no longer predominantly in the calculations of the computers, but the data, their generation, exchange, use and storage take centre stage in a data-centric operation, similar to the products that always form the core of industrial production. “Computer chips and their function are now as complicated as a factory,” explains Professor Rolf Ernst of the Institute of Computer and Network Engineering, or IDA, at TU Braunschweig.
“The core of our research is the self-awareness of the systems, i.e. that given an abundance of data on runtime, they can decide for themselves which data are relevant and needed at a given moment, and which are not. And that they can thus independently adapt themselves and their resource management to the situation without anyone having to actively intervene,” says research associate and doctoral student Nora Sperling, who is involved in the project alongside doctoral students Alex Bendrick and Dominik Stöhrmann in Braunschweig.
The demonstrator at the IDA with its various computers looks less spectacular than the trucks on the university roof in Seoul, but for the Braunschweig team it forms the heart of the work in the IPF 2.0 research project. It consists of four switches, i.e. four “main traffic nodes”. Other computers feed in data from sensors and cameras.
Initially, the focus is on locally networked systems that are still controlled by a common plan, towards dynamically cooperating alliances with emergent behaviour that is comparable to the interaction of factories in global supply chains. Emergent behaviour occurs when the interaction of independent systems gives rise to a higher-level, new function, for example a flexible vehicle network of interacting trucks. Like a factory, an IPF adapts to increasing demands, the availability and state of its resources as well as coordinating the logistics of IPF clusters.
Goals until 2025
The researchers from Germany are focusing on the dynamic control of data-centric vehicle platforms. With “Dynamic NUMA”, a novel data-centric mechanism for data organisation and caching will be developed. This will improve the performance and energy efficiency of distributed systems with high data volumes while enabling compliance with safety and response time guarantees. Among other things, the colleagues from the US are working on being able to replicate driving scenarios in real traffic on the west coast of the United States.
“After the end of the second project phase in 2025, we want to be able to demonstrate that self-awareness will be an important component of the next generation of artificial intelligence,” says Professor Rolf Ernst.
The project “Information Processing Factory 2.0” is a transatlantic cooperation of TU Braunschweig with TU Munich, the University of California in Irvine and the Kookmin University in South Korea’s capital Seoul. The joint funding from the German Research Foundation (DFG) and the US National Science Foundation (NSF) runs until 2025.