bingleilou / fsead

fSEAD: a Composable FPGA-based Streaming Ensemble Anomaly Detection Library

fSEAD

fSEAD: a Composable FPGA-based Streaming Ensemble Anomaly Detection Library

Paper Link: https://dl.acm.org/doi/10.1145/3568992

If you think this code is useful to your project, please cite the paper

@article{lou2023fsead,
  author       = {Binglei Lou, David Boland and Philip HW Leong},
  title        = {fSEAD: a Composable FPGA-based Streaming Ensemble Anomaly Detection Library},
  journal      = {ACM Transactions on Reconfigurable Technology and Systems},
  year         = {2023},
  issn         = {1936-7406},
  volume       = {16},
  number       = {3}
}

---

Introduction

fSEAD is a HLS-based module generator for ensemble-centric streaming anomaly detection. Three detectors are supported: Loda, RS-Hash and xStream.

---

Build

In Ubuntu Linux environment

(1). For environment configuration:

conda env create -f environment.yml
conda activate fsead-test

(2). For params creation: (it spends about 2.5h in a Intel(R) Core(TM) i7-10700F @2.9GHz and 64GB memory PC; optional, for the same parameters used for fSEAD paper, this command is not needed)

bash run_python.sh 

If the gnome-terminal is installed, run the alternative command which is executed in multi-thread terminal. (it spends anout 0.5h)

bash run_python_mt.sh 

(3). For gcc project generation: (the run_gcc.sh works on gcc-7.5.0; the run_gcc11.sh is tested on gcc-11.3.0; the only difference is the '-O3' flag didn't work on gcc11 for the existing project, so this flag is deleted in the run_gcc11.sh)

bash run_gcc.sh 

or

bash run_gcc11.sh 

(4). For hls project generation:(for single thread, run 'run_hls.sh')

bash run_hls.sh 

If the gnome-terminal is installed, run the alternative command which is executed in multi-thread terminal. (7-thread for the existing project)

bash run_hls_mt.sh 

=======

(5). After genreating the HLS IPs, copy the extire path to a windows environment with Vivado v2020.1 installed. The following Vivado project is executated on Windows environment. In the future, we will fix this inconvenience by integrating all the processes in Linux environment.

In Windows environment (For the same BIT/HWH files used for fSEAD paper, this command is not needed) go to fpga/vivado/ in Vivado command terminal

make all

This step spends around 24h in the tested desktop PC.

In PYNQ environment The image version is v2.6

1. delete the original overlay.py and replace it with the fSEAD/fpga/pynq/overlay.py since we modified several bugs of partial reconfiguration function in the official PYNQ image.
2. upload the fSEAD.ipynb, BIT_HWH folder, and dataset to the jupyter environment.
3. execute the fSEAD.ipynb