How we doubled vector search throughput on Elasticsearch Serverless (opens in new tab)

Vector search on serverless was fast until heap copies became the bottleneck. Find out how we gave simdvec direct memory access and doubled throughput.