PofoliaShared via Pofolia

IEEE Computer Architecture Letters· 2025Q2

Cosmos: A CXL-Based Full In-Memory System for Approximate Nearest Neighbor Search

Seoyoung Ko, Hyunjeong Shim, Wanju Doh, Sungmin Yun et al.

Short summary

COSMOS offloads Approximate Nearest Neighbor Search (ANNS) entirely to CXL memory devices, achieving up to 6.72x higher throughput than baseline CXL systems by integrating general-purpose cores and using rank-level parallel distance computation.

AI-generated from the title and abstract; the full text is not read.

Abstract

Retrieval-Augmented Generation (RAG) is crucial for improving the quality of large language models by injecting proper contexts extracted from external sources. RAG requires high-throughput, low-latency Approximate Nearest Neighbor Search (ANNS) over billion-scale vector databases. Conventional DRAM/SSD solutions face capacity/latency limits, whereas specialized hardware or RDMA clusters lack flexibility or incur network overhead. We presentCOSMOS, integrating general-purpose cores within CXL memory devices for full ANNS offload and introducing rank-level parallel distance computation to maximize memory bandwidth. We also propose an adjacency-aware data placement that balances search loads across CXL devices based on inter-cluster proximity. Evaluations on SIFT1B and DEEP1B traces show thatCOSMOSachieves up to 6.72× higher throughput than the baseline CXL system and 2.35× over a state-of-the-art CXL-based solution, demonstrating scalability for RAG pipelines.

The authors' abstract, as published at the source. IEEE Computer Architecture Letters, 2025 · DOI ↗

TakeawaysIn the app
Key pointsIn the app
Ask the paperIn the app

The rest is in the Pofolia app

Takeaways, key points and questions to the paper; new summaries every day for your field. Free.

Sign in on the web to open

Field: Computer Vision and Pattern Recognition

Computer Vision and Pattern RecognitionComputer Science