3 Replies Latest reply on Feb 13, 2015 9:31 AM by Yuriy Fal

    Low performance in Data blending

    daniel lim

      Hi, I am trying to make data blend b/t 2 tables with 2 composite key column (order id, order sub-id. both are in string)

      (Both tables have 300k data rows each.)

       

      But it is extremely slow. it takes 3min to see values in sheet.

       

      The funny things is that both of the table have 'order number' column in number(whole) type, which is not proper column to make join,

      and when I make a data relationship with order number, it just works highly fast.

       

      Does it make sense that data type to be related (either string or number) can influence the performance?

      Or what is the reason that original key column make the blending very slow?

       

      Please help.

       

      Thanks,

      Dan

        • 1. Re: Low performance in Data blending
          Yuriy Fal

          Hi Daniel,

           

          How fast (or slow) a blend is depends on

          cardinality (number of unique values)

          of blended fields. If anything more than 10000,

          the performance degrades significantly.

           

          The type of a blended fields matters, too,

          as number calcs are faster than string ones.

           

          hope this helps.

           

          Yours,

          Yuri

          1 of 1 people found this helpful
          • 2. Re: Low performance in Data blending
            Jonathan Drummey

            In addition to what Yuriy said, there are a few other factors involved:

             

            - Tableau has to query both data sources, get the results, do whatever sorting it needs to do in order to line up the domains, and then do the blend. In this architecture blends will generally be slower than joins.

            - Tableau tries to make this invisible to the end user, but there are two different kinds of data blending, DB1 uses at least 3 queries to get the data, DB2*** uses at least 5, if you're triggering DB2 then that could be slowing things down due to the additional queries involved.

            - Whether your primary data source supports temp tables. Here's an extreme example: In version 8, Tableau Server Published Data Sources that are extracts do not support temp tables until v9, so there have been cases where users saw a view refresh take 5-10 seconds on Desktop and 2-3 _minutes_ on Server. One of the many performance improvements in v9 is that temp table support and Server performance comes much closer to Desktop.

             

            *** Using dimensions from the 2ndary source on any Shelf or having linking dimensions not in the view is what triggers DB2 blending in a particular view.

             

            Jonathan

            • 3. Re: Low performance in Data blending
              Yuriy Fal

              Thank you Jonathan for explaining.

               

              I found blends almost "freeze" on a (v8) server all the time,

              though render acceptably on a desktop.

              Ought to resort to ETL to a single source and outer joins,

              which works but tend to be very time consuming.

               

              Looking (im)patiently for a v9 server release.

               

              Yours,

              Yuri