Best Practices and Design Benchmarks

[SBrown_Murphy]

I'm the tenant admin for our Power BI enviorment and have been compiling a list of best practices and what I call design benchmarks that will be shared with the workspace owners, data engineers, and report developers in order to help them better share the resource we have avalible. 

I'd like to the comunites thoughts on these and suggestions on any that I may have missed.  They are in no perticular order.

Report Design

• Use filters first and slicers only sparingly
• Take away unnecessary visual interactions
• No more then 6-8 visuals on a report page.
• Don't import data that is not actually used to create a visual.
• Aim for DAX queries to complete within 5 seconds.
• Each visual should load within 2-5 seconds.
• Report pages should load in less than 5-10 seconds.
• Better to not use custom visuals, if you must select certified visuals only
• Large ETL work should be performed on the SQL server in a stored procedure or view.
• Large transformations should be avoided in Power BI
• Have a clear objective for the report and limit the number of objectives to the minimum necessary to tell the story.
• Break up large reports into smaller ones and then role them into an app.

Report Data Load

• Data Import should run less than 30 minutes.
• Reports in non-production workspaces should not have a scheduled refresh.
• Large datasets should use incremental refresh
• Avoid unnecessary columns and rows
• Keep the dataset size as small as possible, ideally under 1 GB for optimal performance.
• When scheduling a report refresh look for times with the fewest scheduled refreshes.

Report Publication

• Reports should be published to a Premium Capacity Workspace
• Reports should always use approved gateway connection (never use personal gateways).
• Data Refreshing should be scheduled from within the semantic model. 
• Use a Power BI App to share multiple smaller reports in one place.

Workspace Maintenance

• Using the Power BI Usage Metric report monitor the usage of all your reports monthly.
• Disable refresh schedules for unused reports.
• Archive unused reports in a code repository or on SharePoint.
• Use Active Directory Groups (AD Groups) to manage access to your workspace.
• Limit the number of users in a production workspace with member and contributor roles to as few as possible.
• Workspaces should be targeted to an audience and everything in the workspace should be relevant to that group.
• Reuse datasets whenever the data is the same
• Data Engineers should publish semantic models to their own workspace and then share them with the analysts to be used by reports.

[Anonymous · Accepted solution]

Hi @SBrown_Murphy 

Actually, the description that you post is detailed enough.

I have a suggestion that if you want to share the data to others, maybe you can choose the data flow.

The dataflow can obtain data from the cloud or other applications, which greatly accelerates the transfer of data. For large datasets, it is able to process and clean the data on its own, so it brings a new level of performance and data processing compared to the Semantic Model

Best Regards

Zhengdong Xu
If this post helps, then please consider Accept it as the solution to help the other members find it more quickly.

[SBrown_Murphy]

Thank you for the suggestion.  We don't have anyone using dataflows currently so we will have evaluate them before recommending them as a best practice.  At this point all the report are designed in the Power BI desktop app and then published to a workspace and I don't think anyone has looked at the other objects that are part of pre Fabric Power BI other then symanic modes, reports, dashboards, and apps.  But I will deffenatly look at dataflows.