A Hardware/Software Framework for the Integration of FPGA-based Accelerators into Cloud Computing Infrastructures

The need for high computing power has increased enormously in recent years, particularly in the field of image signal processing and machine learning applications, very powerful computing systems are required. It has been shown that homogeneous architectures in data centers work very inefficiently regarding these special applications, showing high latency of the response times and providing a very poor power efficiency. In order to integrate FPGA (Field Programming Gate Array)- as well as GPU-based accelerators into cloud computing infrastructures as compute nodes we present a generic hardware/software framework for using heterogeneous computing systems. A real industrial image processing application shows the acceleration achieved.