Skip to main content

Power BI

New

Improve SAP BW connector performance and BW memory comsumption

Vote (56) Share
Stephen Fog's profile image

Stephen Fog on 28 May 2018 15:06:12

This idea is really about solving a performance issue with Power
BI‘s memory consumption when getting data from SAP BW.

Basically there are 2 issues but they are closely related, so I mention both in this idea.


1.When connecting to SAP BW, performance is extremely slow and Power BI uses too much SAP memory.

2.When accessing large SAP BW datasets, Power BI continues to use memory , until it consumes all available extended memory, and so impacts other users.


Here is an example of the Power BI memory consumption issue, compared to another SAP tool.



My test case is:
•I compare Power BI with SAP Business Objects Analysis for Office (the Excel Add-in) - both from my desktop.
•I run the same SAP BW/BEx query from PBI and AFO and get the exact same amount of rows and columns.
•The Power BI query has no additional steps (all it contains is Source, Navigation and Added items)
•I deleted the cache from Power BI. I also deleted the database cache in SAP for the BEx query I am using (t/c RSRCACHE).

This is the result from SAP using t/c SM04 and ST02 for both tools, running the same BW/BEx query:

•Power BI, when selecting Edit query and ‘Refresh preview’, takes circa 10 minutes to present the data and consumes more than 500GB of extended memory
•Analysis for office (AFO) takes about 20 seconds to present the data uses 36MB of extended memory – which is less than 10% of what PBI uses)
•Power BI, when you close the query editor and select ‘Apply data’, takes another circa 10 minutes and again consumes 500GB of extended memory.

Comments (5)
Stephen Fog's profile image Profile Picture

9ec20a6e beff-4832-b242-9622e9a8f891 on 16 Aug 2020 04:16:39

RE: Improve SAP BW connector performance and BW memory comsumption

Same issue for us - not only crashes the PBI Desktop refresh, but crashes the auto refresh Gateway process as well. I've created a slight work around by splitting up my BW queries into single year queries and then appending them in the end. It works, but not 100% Have considered splitting it up into six month queries, but definitely a poor solution.

Stephen Fog's profile image Profile Picture

dc580446 aeaa-4f3e-b870-154601197b4b on 16 Aug 2020 04:16:05

RE: Improve SAP BW connector performance and BW memory comsumption

Hi,
We are experiencing the same, 60K rows are being returned in Analysis for Office (Excel plugin) in 20 seconds, the same BEx query when access via Power BI is unable to complete owing to memory availability on the SAP BW server. Many Clients are trying to use power bi with SAP BW however this is a very real limitation.

Stephen Fog's profile image Profile Picture

50e441ed e4dd-4595-9e12-04d3061a820b on 16 Aug 2020 04:12:45

RE: Improve SAP BW connector performance and BW memory comsumption

Got the same issue! PowerBi therefore cannot be used in our (large) organization.

Stephen Fog's profile image Profile Picture

e19d0575 38a5-4da4-a59e-19e40b09e18f on 16 Aug 2020 04:05:52

RE: Improve SAP BW connector performance and BW memory comsumption

Hi,
When using connector 2.0 for SAP BW, when you enter an MDX statement and press the OK button, it will show a preview of the data. It seems that for showing that preview of the data it is not taking in consideration the Batch Size parameter, because it is causing a memory error in BW: "TSV_TNEW_PAGE_ALLOC_FAILED"
Can you please add an option to skip preview, or make sure that preview uses the batch Size parameter.
Thanks!

Stephen Fog's profile image Profile Picture

51f608b8 3b65-4484-9732-728f0b6af166 on 16 Aug 2020 04:04:55

RE: Improve SAP BW connector performance and BW memory comsumption

We are experiencing the same performance issues with the PowerBI BW connector (we are using the new June connector (SAP .net 3.0). The PowerBI connector is taking orders of magnitude longer to run as compared to the same SAP Analysis for Office connector. (e.g. 40 minutes to download 17,000 records). In it's current state we can not use this as a production tool, due to the performance issues.