Effect of DataSet View Calculated Columns on Performance

Domomon

I want confirmation that if I shift over a bunch of dataset view calculated columns to run in the dataflow instead of on the dataset view that I can potentially boost the performance of my dashboards.

Currently, the dashboards run okay, but filtering takes a while. The dashboards are heavily leveraging COUNT(DISTINCT()), which I know is bad for performance, and I hope to eventually develop a workaround that does rely on it quite as much, but for now, confirmation on the effects of a calculated column shift would be appreciated.

GrantSmith

As a view is processed when the data is accessed/visualized in a graph so it can affect the performance of the graphs loading on a dashboard. Moving the logic into a Dataflow will remove the processing of the logic when loading the graphs/dashboard. The caveat to this is that if your dashboard is utilizing filtering your logic won't take the filters into account.

GrantSmith

Beast modes allow you to add conditions to your aggregations dynamically based on what the user selected on the page as filters. Pre-aggregating your data in a Dataflow will set those aggregate values before you're able to do any dynamic filtering as the data is recorded directly into the dataset.

GrantSmith

As a view is processed when the data is accessed/visualized in a graph so it can affect the performance of the graphs loading on a dashboard. Moving the logic into a Dataflow will remove the processing of the logic when loading the graphs/dashboard. The caveat to this is that if your dashboard is utilizing filtering your logic won't take the filters into account.

Domomon

Thanks @GrantSmith! Everything made sense to me up until that last sentence. Can you provide an example or rephrase it?

GrantSmith

Beast modes allow you to add conditions to your aggregations dynamically based on what the user selected on the page as filters. Pre-aggregating your data in a Dataflow will set those aggregate values before you're able to do any dynamic filtering as the data is recorded directly into the dataset.