Ulrich Langenbach

The talk provides an introduction to the corundum project, implementing a 100 GbE NIC based on Commercial-off-the-shelf (COTS) FPGA cards, e.g. Xilinx Alveo Cards. The project consists of all necessary RTL components, e.g. PCIe DMA engine, NIC datapath, MAC, PHY and integration of vendor specific IP cores, such as transceivers and PCIe hard-IP cores. In combination with the provided driver and debugging utility the ready-to-experiment state just requires a supported FPGA card + compiler to kick-off playing with the project.

The existing state provides the platform for future In-Network Compute platform research, allowing for application logic to be balanced across hardware acceleration and software flexibility. The platform uniquely allows for experiments with lower layer protocols, e.g. PHY layer coding.

The talk concludes with an overview about short term and mid term goals of the project.

The OpenSeaMap project is part of the OpenStreetMap project family and aims at leveraging crowd sourced information for leisure captains. As it targets a sepcific user and application group it also relies on additional data on top of traditionally used positional data. This presentation is about an effort to create a hard- and software platform capable of collecting the needed data for the OpenSeaMap project. We share early results yield with a prototype implementation of the proposed platform.

This presentation focusses on the hardware part of a platform prototype used for collecting data on a ship of a participant of the crowd sourced OpenSeaMap project. In combination with the software running on the proposed hardware platform and additional software parts running on the host of an participant and the server side software of the OpenSeaMap project the data is collected, processed, rendered and provided for download and use by the leisure boat captains. The platform is based on a RaspBerry PI and a HAT as well as a professional depth ranger. The project aims at developing a dedicated HAT for the purpose of OpenSeaMap data collection.

However, the current state of development, namely a productively used prototype and the plans for the next steps as well as some early results are shared with the audience. Further ideas and recommendations for improvements and next steps are very welcome.

We share our experiences with a new framework in the Linux kernel for programming reconfigurable devices, namely MPSoC-FPGAs. Our example use case integrates reconfigurable hardware accelerators into the Crypto API. We apply a new, declarative and device-tree-driven reconfiguration framework within the Linux kernel as proposed and implemented by Alan Tull. The implemented concept maps reconfigurable regions within the FPGA to device tree nodes. The insertion of a device tree overlay triggers the reconfiguration of the corresponding reconfigurable region. The reconfiguration process consists of the scheduling, descheduling and execution phase. Based on our usecase, benchmark results for the scheduling phases are shared. We present the bottlenecks revealed by our benchmark and show currently missing components of this approach. We conclude that the current implementation is already in a usable state for developing and deploying MPSoC-FPGA based heterogeneous systems.