NetSHa: In-Network Acceleration of LSH-Based Distributed Search

Locality Sensitive Hashing (LSH) is widely adopted to index similar data in high-dimensional space for approximate nearest neighbor search. Demanding applications (e.g. web search) mean that LSH must exhibit low response times and high throughput. To achieve this, they tend load balance between multiple machines. However, as the scale of concurrent queries volume grow, large numbers messages are required. Hence, network a key bottleneck. address this gap, we propose NetSHa, which exploits computational capacity programmable switches. Specifically, introduce heuristic sort-reduce approach drop potentially poor candidate answers while preserving search quality. Then, NetSHa aggregates good from different when transmitting them. Through it reduces communication cost. Furthermore, best-effort replacement mechanism improve its concurrency. We implement on Barefoot Tofino switch evaluate using 7 real-world datasets. The experimental results show packet by $4\sim 10$ improves efficiency least 3× comparison with typical LSH-based distributed frameworks.