DRIM-ANN: An Approximate Nearest Neighbor Search Engine based on Commercial DRAM-PIMs

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Chen, Mingkai, Han, Tianhua, Liu, Cheng, Liang, Shengwen, Yu, Kuai, Dai, Lei, Yuan, Ziming, Wang, Ying, Zhang, Lei, Li, Huawei, Li, Xiaowei
Format:	Preprint
Published:	2024
Subjects:	Performance
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

_version_	1866912555298979840
author	Chen, Mingkai Han, Tianhua Liu, Cheng Liang, Shengwen Yu, Kuai Dai, Lei Yuan, Ziming Wang, Ying Zhang, Lei Li, Huawei Li, Xiaowei
author_facet	Chen, Mingkai Han, Tianhua Liu, Cheng Liang, Shengwen Yu, Kuai Dai, Lei Yuan, Ziming Wang, Ying Zhang, Lei Li, Huawei Li, Xiaowei
contents	Approximate nearest neighbor search (ANNS) is essential for applications like recommendation systems and retrieval-augmented generation (RAG) but is highly I/O-intensive and memory-demanding. CPUs face I/O bottlenecks, while GPUs are constrained by limited memory. DRAM-based Processing-in-Memory (DRAM-PIM) offers a promising alternative by providing high bandwidth, large memory capacity, and near-data computation. This work introduces DRIM-ANN, the first optimized ANNS engine leveraging UPMEM's DRAM-PIM. While UPMEM scales memory bandwidth and capacity, it suffers from low computing power because of the limited processor embedded in each DRAM bank. To address this, we systematically optimize ANNS approximation configurations and replace expensive squaring operations with lookup tables to align the computing requirements with UPMEM's architecture. Additionally, we propose load-balancing and I/O optimization strategies to maximize parallel processing efficiency. Experimental results show that DRIM-ANN achieves a 2.46x speedup over a 32-thread CPU and up to 2.67x over a GPU when deployed on computationally enhanced PIM platforms.
format	Preprint
id	arxiv_https___arxiv_org_abs_2410_15621
institution	arXiv
publishDate	2024
record_format	arxiv
spellingShingle	DRIM-ANN: An Approximate Nearest Neighbor Search Engine based on Commercial DRAM-PIMs Chen, Mingkai Han, Tianhua Liu, Cheng Liang, Shengwen Yu, Kuai Dai, Lei Yuan, Ziming Wang, Ying Zhang, Lei Li, Huawei Li, Xiaowei Performance Approximate nearest neighbor search (ANNS) is essential for applications like recommendation systems and retrieval-augmented generation (RAG) but is highly I/O-intensive and memory-demanding. CPUs face I/O bottlenecks, while GPUs are constrained by limited memory. DRAM-based Processing-in-Memory (DRAM-PIM) offers a promising alternative by providing high bandwidth, large memory capacity, and near-data computation. This work introduces DRIM-ANN, the first optimized ANNS engine leveraging UPMEM's DRAM-PIM. While UPMEM scales memory bandwidth and capacity, it suffers from low computing power because of the limited processor embedded in each DRAM bank. To address this, we systematically optimize ANNS approximation configurations and replace expensive squaring operations with lookup tables to align the computing requirements with UPMEM's architecture. Additionally, we propose load-balancing and I/O optimization strategies to maximize parallel processing efficiency. Experimental results show that DRIM-ANN achieves a 2.46x speedup over a 32-thread CPU and up to 2.67x over a GPU when deployed on computationally enhanced PIM platforms.
title	DRIM-ANN: An Approximate Nearest Neighbor Search Engine based on Commercial DRAM-PIMs
topic	Performance
url	https://arxiv.org/abs/2410.15621