Skip to main content
cancel
Showing results for 
Search instead for 
Did you mean: 

Earn the coveted Fabric Analytics Engineer certification. 100% off your exam for a limited time only!

Reply
allsopa
Frequent Visitor

slicing and filtering data in reports on either the IPad and through the browser is very slow

We're looking to deploy some PBI reports to our field managers that are connected to a live SSAS tabular instance. However slicing and filtering data in reports on either the IPad and through the browser is very slow – around 7 seconds per interaction

  • The same interactions in the desktop app take about 1 second
  • Trace analysis shows queries against the tabular model take a fraction of a second within the engines, but the overall request duration is 3-4 seconds (in analysing a single request)

In attempting to improve performance, we

  • Reduced the amount of data in the tabular model from 10m to 200k records – no impact
  • Reduced the amount of data being pulled through to the report. i.e. removed the one table that brings through detailed data at the lowest, transactional level – improved by maybe half a second
  • Removed all but 3 visualisations – resulted in 1 second durations, but of course we need all visualisations
  • Removed role security, changed privacy settings, restarted the SSAS connector service (on the same server as the SSAS tabular model), stopped sending usage data – no impact
  • Removed all filters from the filter pane – no impact

 

Can anyone help?

3 REPLIES 3
Greg_Deckler
Super User
Super User

Is the SSAS tabular instance on the same network as the Desktop? If so, it would be an interesting test to put Desktop on a virtual machine in Azure (IaaS) and then see if it suffers from similar performance lag and if so, then reduce the visualizations and see if you get the same performance improvement.  Basically what this will tell you is whether it is something about the Desktop that is just faster than the apps/web or if it is network connectivity between Azure and SSAS tabular instance.

 

It is either that or just the layers of abstraction involved when going from a desktop application to an app/web app. At that point, those things are interacting with web services and the like that the Desktop probably does a lot more efficiently. In an app and web, there is likely a lot of javascript involved, so there's a scripting engine involved versus compiled code. It's going to be slower.

 

I have helped some customers that have had performance issues with many visualizations on a report, in particular, interconnected slicers, where changing one slicer affects all of the other slicers, or a good portion of them. However, the ability to improve this is often directly tied to how things are being done in the model, the relationships behind the scenes, etc. Before delving into that, it would be interesting to rule out network latency.


@ me in replies or I'll lose your thread!!!
Instead of a Kudo, please vote for this idea
Become an expert!: Enterprise DNA
External Tools: MSHGQM
YouTube Channel!: Microsoft Hates Greg
Latest book!:
Mastering Power BI 2nd Edition

DAX is easy, CALCULATE makes DAX hard...

thanks for your replies - apologies for the delay in responding.

i took some additional actions, such as :

  • Loaded the model onto a cloud server and reconnected the reports to this – no impact
  • Rebuilt a report from scratch in case of file corruption – no impact
  • Rebuilt part of the report from scratch using only graphs with multiple values – response time was 1 second
    • Replacing the same graphs with cards resulted in a 3 second response

The conclusion appears to be that PBI (IPad and browser) do not handle single values in cards very well, graphs with nothing on the axis, or treemaps with single values and no groups.
In one report with many cards and single valued treemaps or graphs with no members on the axis (i.e. just 2 measures), changing
• all cards to graphs with multiple members on the axis,
• all graphs to graphs with multiple members on the axis ,
• all treemaps to have multiple members in the groups,
reduced response times from 7 seconds to 2.

 

I'll raise it to the PBI team and see if they can optimise this

 

cheers

greggyb
Resident Rockstar
Resident Rockstar

If you've got snappy responses on desktop and fast execution in profiler traces, then the bottleneck is the round-trip latency and throughput between your datacenter and the Azure datacenter your reports live on.

 

Since minimizing the number of visualizations brought the response time down the most, my guess is that it would be throughput rather than latency as your bottleneck. Hopefully someone will see this who can speak better to how you might go about narrowing down the issue from here.

Helpful resources

Announcements
April AMA free

Microsoft Fabric AMA Livestream

Join us Tuesday, April 09, 9:00 – 10:00 AM PST for a live, expert-led Q&A session on all things Microsoft Fabric!

March Fabric Community Update

Fabric Community Update - March 2024

Find out what's new and trending in the Fabric Community.

Top Solution Authors
Top Kudoed Authors