Funding: BMBF
Project start: 01.09.2022
Project end: 31.08.2025
Project group: Prof. Dr. Wolfgang Karl
Contact:

The research project ScalNEXT - Optimization of data management and control flow of compute nodes for supercomputing is a joint project of JGU Mainz, RWTH Aachen University, TU Munich and KIT with the industrial partner APS Networks funded by the Federal Ministry of Education and Research (BMBF) as part of the funding program "High Performance Computing for the Digital Age 2021-2024 - Research and Inventions for High-Performance Computing" in the field of "New Methods and Technologies for Exascale High Performance Computing".

In order to ensure the scalability of future HPC systems, it is necessary to close the gap in performance between the computing nodes and the network. One way to improve scalability is to outsource control and management tasks to network resources and thus implement a more centralized and scalable organization. To this end, ScalNEXT uses smart networks, i.e. networks that are reconfigurable and programmable. However, to make such smart network components usable, new implementation models are required with implications for operating systems, I/O systems, APIs and applications. In ScalNEXT, the partners are developing and evaluating new technologies at various system levels and for several application areas. ScalNEXT uses runtime environments, operating systems and microkernel approaches to examine the node level of the system. On the other hand, with the transfer of data processing to Smart Switches and the development of new APIs, the switch level is also being considered. Furthermore, the methods developed are applied to the three core application areas of HPC Modeling & Simulation, High Performance Data Analytics, and Machine Learning & Artificial Intelligence.