Hierarchical Filter and Refine using Data Processing Unit

Research Grant
NSF Career Award, CAREER: Communication-efficient and topology-aware designs for geo-spatial analytics on heterogeneous platforms, $511,000, 2022 - 2027. Link

Research Team Member: Dr. Anmol Paudel.

The new research will revisit spatial data analytics on heterogeneous systems comprised of data processing units (DPU). DPU is a next generation of SmartNIC. NIC is Network Interface Card. These DPUs are a new class of programmable processors made by NVidia (and other manufacturers). Similar to a modern smart network interface card, DPUs can be used to filter unnecessary data from overwhelming the CPU and memory bandwidth. DPUs can be used by CPUs to offload computations; thereby reducing the load on CPU and increasing the capability of the compute node.

Two pictures of NVidia Bluefield-2 are shown below.

Traditional CPU-based Filter and Refine is shown below:

DPU-based Hierarchical Filter and Refine is shown below: