Accelerating Recommender Model ETL with a Streaming FPGA-GPU Dataflow

TL;DR

This paper introduces PipeRec, a hardware-accelerated ETL engine using FPGA-GPU dataflows that significantly speeds up data preprocessing for recommender models, reducing training time and operational costs.

Contribution

PipeRec co-designs an FPGA-GPU dataflow-based ETL engine with a training-aware abstraction, enabling high throughput and efficient GPU utilization in recommender system training pipelines.

Findings

ETL throughput increased by over 10x compared to CPU pipelines.

Achieved up to 17x faster ETL than state-of-the-art GPU systems.

Reduced end-to-end training time to less than 10% of CPU-GPU pipelines.

Abstract

The real-time performance of recommender models depends on the continuous integration of massive volumes of new user interaction data into training pipelines. While GPUs have scaled model training throughput, the data preprocessing stage - commonly expressed as Extract-Transform-Load (ETL) pipelines - has emerged as the dominant bottleneck. Production systems often dedicate clusters of CPU servers to support a single GPU node, leading to high operational cost. To address this issue, we present PipeRec, a hardware-accelerated ETL engine co-designed with online recommender model training. PipeRec introduces a training-aware ETL abstraction that exposes freshness, ordering, and batching semantics while compiling software-defined operators into reconfigurable FPGA dataflows and overlaps ETL with GPU training to maximize utilization under I/O constraints. To eliminate CPU bottlenecks, PipeRec implements a format-aware packer that streams training-ready batches directly into GPU memory via P2P DMA transfers, enabling zero-copy ingest and efficient GPU consumption. Our evaluation on three datasets shows that PipeRec accelerates ETL throughput by over 10x compared to CPU-based pipelines and up to 17x over state-of-the-art GPU ETL systems. When integrated with training, PipeRec maintains 64-91% GPU utilization and reduces end-to-end training time to 9.94% of the time taken by CPU-GPU pipelines.