My suggestion is to use a cross data source join that is extracted instead of two extracts in a data blend. I'm thinking this has a good chance to solve your performance issues. The reason why is that a data blend requires at least 2 queries and up to 6 queries across the sources for Tableau to pull the data from each source, identify the linking dimension values and make the blend. What I've found in practice is that once the number of linking dimension values for a data blend gets into the 10s of thousands then data blends start getting too slow for most use cases (such as user interaction via Tableau Server) because of these extra lookups.
By using a cross data source join & extract the resulting data source then all of that lookup will have been done in the extract process and the basic view will only require 1 query to the single extract so it should be nice and fast.
You haven't said how complicated the view/dashboard is, you might be able to get some performance improvement by displaying fewer marks in the initial view and then having users filter & drill down from there, however the data blend might still get in your way.
Two other options for performance would be to set up a custom geography (available now) or to use a custom shapefile (to be available in v10.2). Both would require work outside of Tableau for setting them up but could result in a view that is faster to draw.
Thanks for the response Jonathan. I didn't use a cross data source join as it would be a cartesian join.
My performance data table has data across all regions for 8 products for the past 15 quarters.
So, there are 120 rows for region X in the performance data table.
Region X has 1900 zip codes assigned to it.
Thus, a cross data source join would result in 228k rows for Region X and incorrect results would be calculated.
Can you elaborate on building a custom geography or point to a KB article?
1 of 1 people found this helpful
You can solve this problem by assigning one specific zip code to your data. that means one zipcode contains have data and remaining zip codes having null values. That is how your calculations will not be duplicated.
Sorry for the delayed response. Can you elaborate? I am not understanding what you mean. Thanks