Performance tuning for DirectQuery datasets

[Imagauthamam]

I have prepared a report with my fact table in DirectQuery Mode. Some of the visuals take 10-20 seconds to load. The measure is not complex,it is just a count of a column. If a visual is taking 20 seconds, I want to do performance tuning to make it to load in less than 10 seconds. Where should I start? If someone has worked in DirectQuery Mode please share us your experience.

Right now, what we have done is pushing transformations to the source and created column index

[Imagauthamam]

@amitchandak Thanks Amit for the response. If you have worked on DirectQuery datasets, can you please tell me a rough estimate of how much time an visual takes to load. I know it depends on the dataset. But I just want to understand that the time taken my visual is of acceptable standard or it can be improved further? 

• Our table contains 7 million rows. If I do a count of a column and slice it by a row from dimension table, it takes around 15 seconds. Is it high or is this expected in DirectQuery Mode?

[amitchandak]

@Imagauthamam , make sure joins are on Numeric keys.  Joins are single directional and you have enabled referential integrity