The ubiquitous cameras have produced a deluge of video files. Combining distributed data processing capability of a fast Data Warehouse and the online ML inferencing capability of the Xilinx Alveo FPGA card, Deepgreen DB for Video Discoveries is a platform that enables live tagging of video files.
The platform provides a powerful and convenient standard SQL environment conducive to practical data warehouse operations and, in addition, incorporates primitives that unleash the power of Xilinx Alveo and ML Suite on video files.
Depending on the amount of data/video and query speed required, the Deepgreen DB platform can be scaled out horizontally or vertically by adding compute nodes or Xilinx Alveo resources.
Video files are distributed evenly among the Segment Nodes in the picture above. When the user submits a SQL query, Deepgreen DB splits the query into jobs that it dishes out to the Segment Nodes. Each Segment Node then executes the jobs and processes the video files residing locally in parallel, utilizing the Alveo resources attached to it. This way, a query multiplies its capability in scanning mp4 files by the number of Segment Nodes and Alveo cards available in the cluster.