Server and network infrastructure for a high-performance heterogeneous computing cluster for AI applications

With more than 10,000 students, Bielefeld University of Applied Sciences (FH) is the largest university of applied sciences in Ostwestfalen-Lippe (OWL). With locations in Bielefeld, Minden and Gütersloh, it is excellently networked in the region, nationwide and internationally through diverse contacts, partnerships and cooperations in science, business, politics and culture. High quality in teaching and research is the concern of the departments of Design, Campus Minden, Engineering and Mathematics, Social Work, Economics and Health.

,

Bielefeld University of Applied Sciences sees itself as a scientific institution with a clear application reference and its research profile is oriented towards the global social challenges of the future. In doing so, it places particular emphasis on the demand areas of climate and energy, health, mobility and communication. In all these areas, artificial intelligence (AI) methods are playing an increasingly important role, which is reflected in the numerous ongoing research projects with a strong AI connection at Bielefeld University of Applied Sciences.

,

Limiting factors in the university-wide application and further development of AI methods are the high demand for computing power and data storage and the easy access to the required software, taking into account data protection and IT security. This is where the project "yourAI - Young Researchers Cloud and Edge Computing Platform for AI" (funded by the BMBF in the AI-Nachwuchs@FH funding line) comes in by creating an infrastructure that serves to promote young scientists in the field of AI from bachelor's degree to doctorate and thus has an impact on business, society and science via various channels. In the yourAI project, servers and network infrastructure for a high-performance heterogeneous computing cluster for AI applications are to be procured.

,

The cluster includes storage servers (central file servers and backup servers), virtualization servers (for central applications, services, etc., which are flexibly provided in the form of virtual machines), several compute nodes for use as data nodes in a Hadoop system, several GPU compute nodes (equipped with powerful GPUs for deep learning applications, etc.) and a compute node that is prepared for the installation of FPGA expansion cards. Three networks are planned for the network infrastructure: (1) a high-speed network with at least 200 Gbps, (2) a redundant rack-internal network with at least 1 Gbps, and (3) a network for out-of-band management with at least 1 Gbps.