Data processing facilities
For more than ten years, the Institute of Optical Sensor Systems has been operating a dedicated high-performance computing cluster (HPCC) on its premises. Major use cases for the HPCC installation include:
- Distributed processing of imagery of highly specialized optical and non-optical sensors, for instance, footage of the Modular Aerial Camera System (MACS), into a variety of photogrammetric data products. Outputs comprise digital surface models (DSMs), digital terrain models (DTMs), true-ortho mosaics (TOMs), 3D point clouds, textured mesh geometries and semantically annotated building models.
- Numerical simulation of sensors and related electronic hardware components using custom software applications.
- Distributed interactive and non-interactive 3D rendering and visualization. This includes content creation for project acquisition, immersive illustration of scientific results and geometric validation of photogrammetric algorithms using perfectly accurate synthetic test scenes.
- Software development and portability testing by providing an enterprise-grade Linux platform in addition to the existing personal workstations.
To address these applications, HPCC users currently can access a total of 288 physical central processing units (CPUs) which provide two-way hardware parallelism, i.e., 576 logical CPUs, to cover heavy-duty processing scenarios. The available set of CPUs is distributed among 24 compute nodes, and each node is equipped with 96 GiB of local multi-channel random access memory (RAM). Communication between the compute nodes happens over a high-performance InfiniBand network with a bandwidth of 2 x 40 Gigabit/Second. Since this network is configured redundantly, the loss of processing nodes or one of the two data switches will lead to a gradual degradation in performance, however, the current operation often still can continue.