1Z0-184-25 Question 158
Single answerYour retail analytics team has built a product recommendation engine that uses vector embeddings for real-time search queries. These embeddings are stored in an Oracle Database running on Exadata Cloud Service, which includes Exadata AI Storage. However, your team notices that queries for nearest-neighbor searches are not meeting performance requirements. Which strategy can reliably accelerate these AI vector search workloads on Exadata AI Storage for high concurrency and low latency?
- A
Store embeddings in a traditional B-tree indexed table and rely on Exadata� row-level parallelism.
- B
Implement specialized vector-based indexing and leverage Exadata AI Storage offload features for parallel vector scans.
- C
Load embeddings into an in-memory external cache outside of Exadata to reduce storage I/O overhead.
- D
Partition each embedding table by hash partitioning on user IDs to speed up row lookups.
Show answer and explanation
Correct answer: B
Explanation
Specialized vector indexes designed for similarity search are essential when working with large-scale embeddings. By offloading and parallelizing these similarity operations, Exadata AI Storage can significantly improve throughput and reduce query latency. Refer to Oracle� Exadata documentation and best practices around parallel query offloading and specialized indexing for AI-driven workloads to achieve optimal performance in real-world recommendation systems.
- A. Incorrect.
B-tree indexes are designed for exact match or range queries on scalar data. They are not optimal for vector similarity searches. This approach often results in inefficient lookups when dealing with high-dimensional embeddings.
- B. Correct.
Using specialized vector indexes (e.g., indexes that leverage approximate nearest neighbor algorithms) is the correct approach. Exadata AI Storage can offload and parallelize vector scan operations, significantly reducing latency and improving throughput for real-time recommendations.
- C. Incorrect.
While caching might offer benefits, it does not take advantage of Exadata� integrated AI Storage capabilities. Managing and synchronizing external caches can also introduce complexity and potential data inconsistency.
- D. Incorrect.
Hash partitioning based on user IDs can help distribute data evenly but does not directly accelerate vector similarity computation. For vector search, specialized indexes combined with Exadata's AI offload features generally yield better performance.