1Z0-184-25 Question 155
Single answerYou are working with a media-streaming service that uses user preference embeddings to recommend new content. You have moved these vector embeddings to an Oracle Exadata environment with AI Storage to handle high-performance AI vector searches on large datasets. However, your similarity search queries still experience latency under peak loads. Which approach should you implement to fully leverage Exadata AI Storage and maximize query performance?
- A
Partition the embedding data across multiple table spaces and rely on standard B-tree indexing to evenly distribute load.
- B
Use specialized vector indexing for approximate nearest neighbor searches and store the indexes on Exadata AI Storage.
- C
Convert all embedding data to JSON documents and use Oracle text-based indexing for vector similarity queries.
- D
Store only vector IDs in Exadata and the actual vectors in local ephemeral disk to minimize data retrieval overhead.
Show answer and explanation
Correct answer: B
Explanation
To accelerate AI vector searches on Oracle Exadata AI Storage, the recommended practice is to use specialized vector indexing (e.g., ANN indexes) that can efficiently handle high-dimensional similarity lookups. Standard partitioning or text-based indexes do not address the core needs of vector search, and storing vectors outside Exadata can undermine both performance and consistency. Oracle� best practices (as referenced in official documentation for vector data management) emphasize building and maintaining vector indexes within the database, leveraging Exadata AI Storage for optimal throughput and scalability.
- A. Incorrect.
Incorrect. While partitioning helps with data distribution, B-tree indexes are not optimized for vector similarity queries. You need a vector-oriented indexing approach to search embeddings efficiently.
- B. Correct.
Correct. Specialized vector indexes for approximate nearest neighbor (ANN) searches are required to handle high-performance similarity queries at scale. Exadata AI Storage can accelerate these searches by optimizing how the indexes are stored and accessed.
- C. Incorrect.
Incorrect. Converting embeddings to JSON and relying on text-based indexing is not well-suited for multi-dimensional vector queries. This approach can significantly degrade performance when dealing with large, high-dimensional embeddings.
- D. Incorrect.
Incorrect. Storing vectors separately on ephemeral storage risks data consistency and does not leverage the benefits of Exadata AI Storage's built-in optimizations for vector indexing.