ERBium Business Rule Engine Hardware Accelerator

Compatibility

XRT 2.3.1301
Vitis 2019.2
Ubuntu 18.04.3 LTS

Board	Shell
Xilinx Alveo U250	xilinx_u250_xdma_201830_2
Xilinx Alveo U250	xilinx_u250_qdma_201920_1
Xilinx Alveo U280	xilinx_u280_xdma_201920_1
AWS F1 VU9P	xilinx_aws-vu9p-f1_shell-v04261818_201920_1

NFA Generation

Two files are required;

Rules data set, formatted as data/demo_01.csv; and
Rule structure file, formatted as data/demo_ruletype.csv.

cd sw
make <demo|benchmarks>

For advanced compilation, check file sw/erbium.cc and read all parameters.

Place sw/<build_dir>/cfg_criteria_<heuristic>.vhd into ./hw/custom/

Hardware Compilation

Set the COMMON_DIR variable in file xilinx_work/Makefile to your local Xilinx Vitis Accel Examples directory.

cd xilinx_work
make <all|build|compile> TARGET=<sw_emu|hw_emu|hw> DEVICE=<FPGA Platform>

AWS F1 Generation

You first need to setup the environment and install the AWS F1 shell. More information here.

cd ..
git clone https://github.com/aws/aws-fpga.git $AWS_FPGA_REPO_DIR
cd $AWS_FPGA_REPO_DIR
source vitis_setup.sh

The following will take care of generating the bitstream, the AFI and notification script:

cd erbium/xilinx_work
make aws_notify TARGET=hw DEVICE=$AWS_PLATFORM EMAIL=<your.email@for_nofication_when.done>

Execution

cd erbium/xilinx_work
make <check|run> TARGET=hw DEVICE=<FPGA Platform>

For advanced execution, check file sw/kernel_<shell>.cpp and read all parameters.

Publications

Fabio Maschi, Muhsen Owaida, Gustavo Alonso, Matteo Casalino, and Anthony Hock-Koon. Making Search Engines Faster by Lowering the Cost of Querying Business Rules Through FPGAs. In Proceedings of the 2020 ACM SIGMOD International Conference on Management of Data (SIGMOD’20), June 14–19, 2020, Portland, OR, USA. ACM, New York, NY, USA, 16 pages. https://doi.org/10.1145/3318464.3386133 [Publication] [Paper] [Talk] [Slides] [Code]

Citation

If you find the ERBium engine useful, please consider citing the following paper:

@inproceedings{maschi2020erbium,
  author    = {Fabio Maschi and
               Muhsen Owaida and
               Gustavo Alonso and
               Matteo Casalino and
               Anthony Hock{-}Koon},
  title     = {Making Search Engines Faster by Lowering the Cost of Querying Business
               Rules Through FPGAs},
  booktitle = {Proceedings of the 2020 International Conference on Management of
               Data, {SIGMOD'20}, June 14-19, 2020, Portland, OR, USA},
  pages     = {2255--2270},
  publisher = {{ACM}},
  year      = {2020},
  url       = {https://doi.org/10.1145/3318464.3386133},
  doi       = {10.1145/3318464.3386133}
}

fpgasystems/erbium