SmartSSD GNN Sampling – Code Repository Overview

Table of Contents

SmartSSD-based GNN sampling implementation accompanying the paper “In situ neighborhood sampling for large-scale GNN training.” It runs sampling near storage on SmartSSD FPGAs to cut data movement and speed up training on large graphs.

Repo: https://github.com/CASP-Systems-BU/gnn-sampling-with-smartssd

Features #

Streaming sampler (FPGA-accelerated, multi-layer)
Random-read sampler (alternative access pattern)
Multi-device support
P2P transfers (SSD ↔ FPGA, no host memory)

Quickstart #

Requirements: SmartSSD, Vitis 2021.2, XRT, CMake, GCC, OpenMP
Env: source Vitis and XRT scripts
Build: CMake in build/ then make -j
Run: ./test_streaming_sampler

Reference #

DOI: https://doi.org/10.1145/3662010.3663443
License: See repository LICENSE