Institute for Data Processing and Electronics (IPE)
Area of research:
Cloud computing is a convenient tool that enables ubiquitous access to a shared pool of configurable system resources. Cloud technologies allows scientists to get their applications up and running faster, with improved availability and less maintenance, and that it enables IT teams to more rapidly adjust resources to meet fluctuating and unpredictable demand. OpenShift is a Software as a Service (SaaS) cloud platform from RedHat for container-based deployment and management. Up to now SaaS platforms are mainly used to provide services using general-purpose processors only. But GPU computing has become an essential part of scientific imaging and data analysis platforms and are used to process the ever-growing amounts of data recorded by scientific instrumentation world-wide.The goal of this Master thesis is to evaluate the applicability of the OpenShift platform for GPU workloads. System stability and resource isolation should be evaluated. The student is expected to characterize all possible sources of additional latencies: What are the additional kernel launch latencies? How big are the performance penalties due to transfer of data between GPU and system memory? Can advanced techniques like GPUDirect and pinned memory be used to speedup data transfer? Finally, it is important to evaluate how well GPUs can be shared by the containerized applications and the optimal number of applications per GPU unit should be suggested. Upon successful evaluation, a test implementation of our 3D visualization platform should be realized.
Good background in systems engineering, strong knowledge of C/C++ programming languages. Prior experience in parallel programming of GPUs using CUDA or OpenCL is a plus.
According to the study regulations
Dr. Suren Chilingaryan, IPE, phone: 0721 608 26579 (firstname.lastname@example.org)
Dr. Andreas Kopmann, IPE, phone: 0721 608 24910 (Andreas.email@example.com)