1Z0-184-25 Question 161
Single answerYou are designing a recommendation engine in Oracle Cloud Infrastructure using Oracle Exadata Cloud Service with Exadata AI Storage to accelerate vector-based queries on millions of product embeddings. To achieve fast nearest neighbor searches, you need to ensure that query processing can be offloaded to Exadata AI Storage. Which approach should you implement when storing and indexing the vector data?
- A
Create specialized approximate nearest neighbor indexes on the vector columns, supported by Oracle Database, so that Exadata can offload vector search.
- B
Store the vector data as raw text in CLOB columns, and rely entirely on B-tree indexes and filtering.
- C
Use partial indexing on each numerical dimension of the vectors separately to handle approximate nearest neighbor queries.
- D
Migrate the vector embeddings to Oracle NoSQL Database Cloud Service and replace Exadata AI Storage with a separate search solution.
Show answer and explanation
Correct answer: A
Explanation
Exadata AI Storage can accelerate AI workloads, including vector-based nearest neighbor searches, by offloading intensive operations to specialized storage cells. To utilize this offload effectively, you must store and index vectors in a manner that Oracle Database can process natively with approximate nearest neighbor methods. Storing vector data in CLOBs or splitting them into multiple numeric columns prevents Exadata's specialized offload from being used effectively. For more details, refer to Oracle Database documentation on advanced indexing and Exadata offload features.
- A. Correct.
Option 1 is correct. By creating approximate nearest neighbor (ANN) indexes on your vector data, you can leverage the Exadata AI Storage offload capabilities for vector similarity searches. Oracle Database supports specialized indexing for vectors that significantly speeds up searches, especially on Exadata.
- B. Incorrect.
Option 2 is incorrect. Storing vectors in CLOB columns and relying on B-tree indexing will not effectively leverage Exadata AI Storage for vector offload. B-tree indexes are not optimized for similarity or distance-based searches on high-dimensional data.
- C. Incorrect.
Option 3 is incorrect. Indexing each dimension separately (e.g., one numeric column per dimension) is cumbersome and does not provide the efficient distance-based capabilities needed for high-dimensional vector search. It also does not effectively offload to Exadata AI Storage.
- D. Incorrect.
Option 4 is incorrect. Migrating the embeddings to Oracle NoSQL Database Cloud Service does not utilize Exadata AI Storage's specialized offload mechanisms. This approach moves your data away from the performance enhancements provided by Exadata.