1Z0-184-25 Question 156
Single answerYour e-commerce application hosts millions of product embeddings for real-time recommendations. You want to leverage Exadata AI Storage in Oracle Cloud Infrastructure to accelerate vector-based searches (approximate nearest neighbor). Which deployment strategy provides the best balance of high concurrency and low query latency for these AI workloads?
- A
Store all embeddings in a single, non-partitioned table and rely exclusively on full table scans in parallel.
- B
Partition embeddings by product category and create vector indexes on each partition, leveraging Exadata� parallel query engine.
- C
Use external object storage for embeddings, loading them into Exadata AI Storage only when a user request arrives.
- D
Disable auto-tiering within Exadata AI Storage so that all vector data resides in memory across all Exadata storage servers.
Show answer and explanation
Correct answer: B
Explanation
The key to accelerating AI vector searches on Exadata AI Storage is to organize data for efficient parallel execution. In this scenario, partitioning embeddings by category and creating specialized vector indexes balances concurrency and speed, leveraging Exadata's parallel query engine. For more details, refer to Oracle Documentation on vector indexing and Exadata best practices, including the Oracle Database 23c (and later) features for approximate nearest neighbor searches.
- A. Incorrect.
Option 1 is incorrect. Relying solely on full table scans in a single, non-partitioned table can slow down queries under heavy load. Parallel scans help but are not as efficient as specialized vector indexes combined with partitioning.
- B. Correct.
Option 2 is correct. Partitioning the embeddings by category and creating vector indexes allows for parallelizing searches at the storage layer. Exadata AI Storage can process different partitions in parallel, enhancing concurrency and query performance for vector-based workloads.
- C. Incorrect.
Option 3 is incorrect. Continuously loading data from external object storage introduces substantial latency and overhead, undermining any performance gains from Exadata� specialized AI features. Storing the embeddings within Exadata is more efficient for frequent searches.
- D. Incorrect.
Option 4 is incorrect. Simply disabling auto-tiering to keep data in memory does not address the need for organized vector lookup (via indexing and partitioning). Without indexing structures, memory residency alone will not ensure optimal query performance under load.