4x Performance Boost with Elasticsearch v9 Upgrade for Vector Search
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
Trendyol ran a head-to-head load test comparing Elasticsearch v9.2.1 against v8.18.3 on a 27-node cluster with 9 million documents and dual dense vectors (415d + 256d). Using hybrid search queries (filter + k-NN + script scoring), v9.2.1 delivered roughly 2× the throughput, ~50% less CPU usage, and up to ~10× better p99 latency under high concurrency. At 960 concurrent workers, v8's p99 latency hit 1.95 seconds while v9 held at 188 ms. Beyond 960 workers, v8 crashed entirely while v9 scaled to 40k+ req/min. The conclusion is a strong recommendation to upgrade from 8.x to 9.2.1 for any vector or hybrid search workload.
Table of contents
Key takeaways (at a glance)Introduction🏗 1. The lab: infrastructure and stress scenarioSort: