1Z0-184-25 Question 157
Single answerYou are implementing a semantic search feature for an enterprise knowledge base using Oracle Database on Exadata, where millions of text documents are converted into multidimensional vectors for similarity searches. You want to leverage Exadata AI Storage to speed up approximate nearest neighbor (ANN) queries for these vectors. Which configuration best achieves this goal?
- A
Store the vectors in a B-Tree index to maximize Exadata Smart Scan performance.
- B
Create a specialized vector index on the embedding column and enable offload processing to Exadata AI Storage.
- C
Replicate all vectors to application servers� local memory to minimize data movement from Exadata.
- D
Use separate columns for each dimension of the vectors for improved partition pruning.
Show answer and explanation
Correct answer: B
Explanation
When performing AI vector searches on Exadata, using Oracle� specialized vector data type and approximate nearest neighbor indexing allows Exadata AI Storage to offload and parallelize scanning operations. This offload mechanism significantly reduces the query processing load on the database servers. Refer to Oracle� documentation on AI vector search best practices and Exadata offload processing guidelines for more details on how to configure indexing and data placement to optimize performance.
- A. Incorrect.
Option 1 is incorrect because B-Tree indexes are not well-suited for high-dimensional embedding data, and Exadata Smart Scan offers the best benefits when used with appropriate data types and indexing methods designed for vector-based queries.
- B. Correct.
Option 2 is correct. Oracle Database supports specialized vector indexes for ANN queries, and Exadata AI Storage offloads the heavy scan operations to the storage layer. This configuration provides the most efficient vector search by minimizing network overhead and parallelizing computations.
- C. Incorrect.
Option 3 is incorrect because replicating large embeddings to all application servers is typically inefficient and does not leverage Exadata AI offload capabilities. It also introduces significant overhead in terms of memory usage and data synchronization.
- D. Incorrect.
Option 4 is incorrect as storing each dimension in a separate column makes partition pruning less effective for vector queries and complicates indexing. Specialized vector indexing is the recommended approach to handle high-dimensional embeddings.