Dataset size limits and performance

This topic describes the dataset size limits and expected performance of each Arria for Power BI feature.

Custom Narratives

The size of allowed datasets has increased in the current release. Previous versions of Arria for Power BI had a dataset limit of 30,000 rows. Improvements to the add-in mean that the number of rows in your dataset is no longer limited. Now, the only restriction is the data payload limit imposed by Studio (20MB). (The data payload is the subset of fields you select from your original dataset to be sent to the Custom Narratives service).

Various factors affect the time it takes to generate custom narratives — mainly network speed, the complexity of the Studio project, and the size of the data payload sent to the Custom Narratives service.

For example, assuming download/upload speeds of 50MBps/8MBps:

  • generating a custom narrative from a dataset of around 8,000 rows and 8 columns (data payload approx. 1MB), may take up to 2 seconds.

  • generating a custom narrative from a dataset of around 100,000 rows and 8 columns (data payload approx. 10MB), may take up to 16 seconds.

NLG Apps

NLG Apps generate narratives describing a subset of dimensions and measures (data fields) that you select from your data. The more targeted the dataset is, the more pertinent the narrative will be.

NLG Apps perform best when the size of the dataset sent to the NLG Apps service (i.e., the data payload) is around 2MB or less.

For these reasons, we recommend that you select only those dimensions and measures required for the particular narrative you wish to generate. For example, for a time-based variance analysis of sales drilled down by product and segment, you might select one temporal dimension, plus your sales, product, and segment measures. This will result in a smaller aggregated dataset.

Arria Answers

Arria Answers allows you to gain insights into a subset of dimensions and measures (data fields) that you select from your data, according to the questions you ask.

Arria Answers performs best when the size of the dataset sent to the Arria Answers service (i.e., the data payload) is around 5MB or less. Performance will vary based on dataset size.

For this reason, we recommend that you select only those dimensions and measures required to answer your queries.