Cheapest store for 60 million rarely searched embeddings
MLA-C02
Data Preparation for ML and AI
Medium
Question 26 of 30
A legal research firm keeps 60 million vector embeddings of historical case documents for a semantic search tool. Researchers run a few hundred searches each day, and results within one second are acceptable. The current vector database runs on provisioned cluster nodes that sit idle most of the day. The firm wants to lower storage and query costs and to stop managing vector database infrastructure.
Which solution will meet these requirements MOST cost-effectively?
Show answer and explanation
DMove the embeddings to an Amazon S3 vector bucket and query them through an S3 Vectors vector index
Amazon S3 Vectors meets the requirements. Vector buckets store and query vectors without any infrastructure to provision, the firm pays only for what it uses, and AWS describes subsecond query latency for infrequent queries, which fits a few hundred searches a day. An OpenSearch Service managed cluster with larger nodes adds provisioned capacity that sits idle and still leaves a cluster to manage. Computing similarity in Lambda means each search reads and compares millions of stored vectors, which is neither fast nor cheap. An RDS for PostgreSQL instance sized for the index is provisioned compute that runs all day, and it is still a database to operate.
Source: docs.aws.amazon.com/
Anonymous, no account needed. Your tally stays in this browser.
Practice by topic: RAG questions