SKT - A One-Pass Multi-Sketch

SKT is a C++-implemented HLS kernel for computing multiple sketches (HyperLogLog, Fast-AGMS and Count-Min) over the data input in one pass. This complete set is accommodated on a single Alveo U250 accelerator card operated under the Xilinx QDMA shell without backpressure on the PCIe data link. A detailed anylysis of possible design trade-offs in terms of sketch dimensions and input bandwidth will be shared shortly.

Repo Structure

• hw - the SKT hardware accelerator kernel + example host application
• sw - SW-SKT, the SKT software reference implementation