Explore a cutting-edge approach to billion-scale approximate nearest neighbor search (ANNS) in this conference talk from USENIX ATC '23. Dive into CXL-ANNS, a software-hardware collaborative solution that leverages Compute Express Link (CXL) technology to disaggregate DRAM and enable highly scalable ANNS services. Learn how this innovative method addresses performance challenges by caching frequently visited neighbors, implementing intelligent prefetching, and maximizing search parallelism across the CXL interconnect network. Discover the impressive performance gains achieved by CXL-ANNS, including 111.1x higher QPS and 93.3% lower query latency compared to state-of-the-art ANNS platforms, as well as its superiority over DRAM-only systems in terms of latency and throughput.
Overview
Syllabus
USENIX ATC '23 - CXL-ANNS: Software-Hardware Collaborative Memory Disaggregation and Computation...
Taught by
USENIX