Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Learn about Quake, an innovative adaptive indexing system designed to optimize vector search performance in dynamic environments through this 15-minute conference presentation from OSDI '25. Discover how this system addresses the limitations of existing approximate nearest neighbor (ANN) methods that struggle with dynamic and skewed workloads where data distributions continuously evolve. Explore Quake's multi-level partitioning scheme that automatically adjusts to updates and changing access patterns, guided by a sophisticated cost model that predicts query latency based on partition sizes and access frequencies. Understand how the system dynamically configures query execution parameters to meet recall targets using a novel recall estimation model, and examine its NUMA-aware intra-query parallelism approach for enhanced memory bandwidth utilization during search operations. Gain insights into the comprehensive evaluation methodology, including a Wikipedia vector search workload and configurable workload generator, that demonstrates Quake's superior performance with query latency reductions of 1.5–38× and update latency reductions of 4.5–126× compared to state-of-the-art indexes like SVS, DiskANN, HNSW, and SCANN, making it particularly valuable for machine learning applications including retrieval-augmented generation, recommendation systems, and information retrieval.
