auto, navigation, gps, vehicle, dashboard, interior, transport, map, gps, gps, gps, gps, gps
Photo by USA-Reiseblogger on Pixabay

Dashboard Design

Part of BI system performance and cost control

Diagnosing slow dashboards before replacing the tool

Reproduce a slow dashboard action, locate the delay, and test whether a targeted fix meets the reader’s need before considering replacement.

This is an editorial pen name used by the publication, not a claim of an independently verified personal identity. Original editorial credit: Desk Analytics.

Before replacing a slow dashboard tool, reproduce the delayed action and locate where the waiting occurs. Separate visual rendering from model calculation, source queries, refresh, and gateway or network effects. A replacement that runs the same slow source query may inherit the delay.

Make the complaint repeatable

Ask whether the problem is first page load, a filter change, a drill into detail, or a refresh that finishes too late. Record the report and page, filter state, approximate time, user role, and whether the figures were current. Measure the action where readers encounter it.

Keep the data snapshot, device, connection, and report version comparable where practical. Record whether the report or source result was already cached. If the delay appears only when colleagues are active, include a representative busy period as well as a quiet one.

Split the wait

ObservationNext inspectionQualification
One visual is slowIts query, calculation and display timeIt may also wait for other page work.
Every visual slows after a filterSource and model work triggered by that interactionConfirm which queries reach the source.
Page is quick but figures are oldSource load, model refresh and visible cutoffA new render does not prove current data.
Delay appears at busy timesConcurrent queries, refreshes and capacityA quiet test may miss contention.

In Power BI, Performance Analyzer records visual load durations after interactions while editing a report, and it breaks each load time into categories, such as the time the DAX query took to run. It can help identify which visual is affecting report performance and which aspects are taking the longest duration.

For the refresh layer, open the semantic model's refresh history in the Power BI service and switch to the Scheduled tab to see scheduled data refreshes. Compare their timing with the busy period and with the cutoff readers expect.

For the gateway layer, on-premises data gateway performance logging is turned on by default, and Microsoft offers a Gateway Performance PBI template file to visualise the query logs. When investigating slow queries, you can temporarily enable Additional logging to capture query strings, mashup engine logs and verbose-level tracing.

For contention on shared compute, the Microsoft Fabric Capacity Metrics app monitors capacity consumption and can help identify when to scale up a capacity or turn on autoscale. Check it for the same busy period before blaming the report.

In Tableau Server, a performance recording gives the same split. Its Performance Summary dashboard sorts events by duration, including Connecting to data source, Compiling query, Executing query and Computing layouts. Tableau notes that slow connections could be due to network issues or the database server, while long layout times point to the workbook.

Change the identified cause

If a dense page is responsible, move detail to another view. If a source query reads unnecessary data, narrow it without changing the approved population. If model calculation is slow, review the calculation and relationships with its owner. If refresh overlaps interactive use, examine the schedule before adding capacity.

Compare the old and revised headline figure, useful breakdown, cutoff, and access for representative users. Measure the same action again under comparable conditions. A shorter wait is useful only if the report still answers the approved question.

Record the result per layer, in seconds, before and after. For example, if most of a month-end page's wait sits in one visual's DAX query time, simplify that measure and time the same filter state at the same busy period, then confirm the totals and access are unchanged.

Decide whether replacement is justified

Set an acceptance case for the action, audience, permitted data, expected figure, freshness, and acceptable wait under a stated workload, stated as a number of seconds for a named action at the month-end peak and agreed with the report owner. Consider replacement when feasible changes to the current arrangement cannot meet that need. Apply the same case to the candidate.

Keep tuning while the measured wait sits in a layer you control: a visual, a calculation, a refresh schedule, the gateway or capacity. Replacement is justified only when the remaining wait sits in the tool itself and the candidate meets the acceptance case on the same data, filters and workload.

If most of the wait is the source query, a new tool reading the same source is likely to inherit it. A demonstration using a smaller extract or different filters does not establish that it will solve the reported delay.

Further reading: Use Performance Analyzer to examine report performance (learn.microsoft.com).

More from Dashboard Design