Optimize Dataset Performance in Oracle Analytics DV

Dhaval Parikh Stantec

What we have observed is that Oracle Analytics DV datasets created from subject areas are often significantly slower than querying the subject areas directly, even when they contain fewer columns and less data. This performance gap affects user adoption, trust in the tool, and the efficiency of self-serve analytics. Some of the issues we think might be affecting the performance of slower datasets:

1. Subject areas leverage BI Server query optimizations, while DV-generated datasets often do not, resulting in suboptimal SQL execution.
2. DV datasets does not inherit query optimization techniques from the BI subject area model instead of treating them as separate queries.
3. Joins and aggregations in datasets are often recomputed instead of being optimized like in subject areas.

Can some enhancements be made to improve the performance of datasets such as ensuring that the datasets leverage BI Server query optimizations – Queries executed in datasets should be as optimized as subject area queries, inheriting join strategies, indexes, and aggregations, implementing smarter push down processing to the database instead of computing within DV or enhancing SQL Generation & Execution plans?

Rajakumar Burra

Yes. 100% agree with this issue.

SAs are always faster than datasets.

Due to this we are not using the datasets created from SAs.

We are using datasets created from SAs in Dataflows and Dataflow using output in Workbook.

Bret Grinslade - Oracle Analytics-Oracle

@Dhaval Parikh Stantec - do you have some additional context here? I assume you mean creating an LSA. Overall LSAs are very similar to just using an SA directly. Perhaps you can explain an example where you see a different in performance (such as adding joins, joining across sources, etc.)