As AI agents become the primary interface for data retrieval, triggering numerous parallel queries in rapid succession as they reason and iterate, search throughput demands increase dramatically. This challenges traditional REST-based JSON serialization, where data movement and computation can consume up to 60–80% of query execution time at scale. In addition, search engines are increasingly converging with analytical data systems, which drives adoption of columnar data layouts, SIMD-accelerated execution, and bulk-oriented streaming data pipelines.
This session showcases how OpenSearch, an open source top-level project under The Linux Foundation, integrated Apache Arrow and Arrow Flight to achieve 10x performance improvements for analytical queries. We will walk through the streaming transport implementation using Arrow Flight RPC for zero-copy data transfer; how columnar storage structures improve cache locality and enable efficient evaluation of aggregations and filters; how SIMD and vectorized computation transform scoring, filtering, and similarity operations by leveraging modern CPU capabilities; and how bulk ingestion and execution pipelines reduce coordination overhead while enabling deeper hardware utilization.
Furthermore, we’ll explore how adopting Arrow opened up exciting architectural options in OpenSearch with the ASF ecosystem: we’ll show the integration with Apache Kafka for an end-to-end streaming ingestion architecture with zero-copy deserialization on the consumer side. And we’ll discuss how the adoption of Apache Parquet makes a perfect match, with Parquet optimized for storage and Arrow optimized for compute. More importantly, we will discuss the design trade-offs, integration challenges, and lessons learned while bringing these systems together.
This collaboration between OpenSearch and ASF projects exemplifies the power of open source. Hear from OpenSearch Ambassadors on the latest innovation from OpenSearch V3, as well as the design decisions and the learnings from adopting these popular ASF projects within the OpenSearch project.
