IMPORTANT NOTE: Timing of sessions and room locations are subject to change.
The Sched app allows you to build your schedule but is not a substitute for your event registration. In order to attend OpenSearchCon Europe 2026, please visit our website to register.
This schedule is automatically displayed in Central European Summer Time (UTC+02:00). To see the schedule in your preferred timezone, please select from the drop-down located at the bottom of the menu to the right.
Sign up or log in to add sessions to your schedule and sync them to your phone or calendar.
Single Instruction, Multiple Data (SIMD) is a key feature of modern CPUs, allowing a single instruction to process multiple data elements in parallel. For large-scale vector search workloads, fast dot-product computation is crucial for low-latency, high-throughput semantic search.
In this talk, we’ll share our experience accelerating Lucene’s vector search using native SIMD on ARM CPUs (AWS Graviton 2/3). We’ll first discuss challenges with Java's native access improvements (Project Panama) for vectorization, and why we switched to native SIMD implementations for better performance. Through architecture-specific optimizations, we reduced vector search latency on ARM-based systems by 60%.
Next, we’ll dive into the complexities of working with multiple CPU families, each with different SIMD instructions and optimizations. We’ll also share our progress on making this optimization available to all Lucene users, carefully iterating with Lucene developers to address concerns about linking native code to Lucene’s core.
Finally, we’ll highlight our efforts to make this SIMD optimization an opt-in feature and its potential impact on OpenSearch deployments on ARM at scale.
Shubham is an Apache Lucene Committer and Software Engineer at Amazon Search, working on building and scaling the core search engine powering Amazon's product search. He joined Amazon in 2020 and has nearly 5 years of experience developing large-scale information retrieval systems... Read More →
