Forum Discussion
Power BI Desktop import speed from Power BI Dataflows (ADLS2)
Hi Ibendlin,
I love your term "glorified CVS files", I have certainly not seen this term anywhere in Microsoft marketing materials.
In our organization we have been using PBI Dataflows since the day we rolled-out Power BI, that was approx. 3 years ago. The main selling point of DFL was the simplicity of provisioning data to our users from various systems and so far with mostly positive feedback. We currently have around 1500 dataflows across our capacities.
Without having the deep technical knowledge, I would guess that there should not be performance effect of our Gen2 capacities in the import speed from dataflows. Import in this case is simple read operation on the ADLS side.
As you have suggested, "...some weird column conversions in some of your dataflows..." is also where my money is, but I would like to understand more about this behaviour, which was the reason to post here - to fish ideas from others 🙂
Helper IIIIf you store your dataflow results into a dedicated ADLG2 instance you'll see that they are indeed snapshotted CSVs with no header information. The column names and types are found in the Common Data Model file of model.json at the top level of the data flow.
If I download one of our larger dataflow CSVs (2.55 GiB) using Microsoft Azure Storage Explorer it speeds along at an average of ~700 Mbps which is significantly better than the ~20Mbps we see when pulling in the data via PowerBI Desktop. 🙄