8 Replies Latest reply on Feb 16, 2016 10:19 AM by Eric Axelrod

    Publishing extracts is slow (performance issues)

    Jon Cantrell

      Hello, I'm working with pulling data from Project Server to Tableau. The way we have to slice our data makes live connections very slow. Our users however, are demanding data on a refresh schedule that our Tableau Server admins aren't comfortable with (10-15 minutes) so we can't schedule extracts in Tableau. We've been looking at Alteryx as a solution to fill this gap and extend capabilities of blending data. The problem, however, is when publishing larger extracts to the Tableau Server. It takes 20+ minutes to upload a 120MB extract. I have a workbook in Tableau that has a different extract of about 70MB and that takes 15+ minutes to publish. It seems that bandwidth isn't the constraint. Is there something that can be done or configured on the Tableau Server to improve the throughput when publishing an extract?

       

      Thanks,

      Jon

        • 1. Re: Publishing extracts is slow (performance issues)
          Yuriy Fal

          Hi Jon,

           

          It is considered to check for the following things:

           

          1) Check if your TS disk subsystem would be fast enough.

          It is wise to have a RAID-10 of SAS-6 / SAS-12 HDDs,

          or even use server-grade SSDs if you can afford it.

          Typically, you may have IOPS at above 1000.

           

          2) Consider having dedicated node(s) in your TS cluster

          for Data Engine process(es) together with API Server process --

          and may be Data Server / Cache Server / Backgrounder ones as well.

           

          The main reason for this is to exclude a concurrency for node cores 

          between VizQL-heavy transactions and Data Engine-heavy ones.

           

          3) One last thought, and it is very hypothetical:

           

          Since each copy of an extract is uploaded and published to TS

          via Alteryx-initiated REST API call (series of calls, actually,

          if you uploading a file larger than 64MB) one could test a possibility

          to publish the "same" extract to TS (overwriting the one already there)

          in a somewhat "overlapping" mode -- starting the next upload/publish task

          in another Alteryx flow before the current one is finished.

           

          This would create a "que" of upload/publish REST API transactions,

          and if Tableau Server has enough muscles to sustain such a workload,

          those transcations could be initiated with as short time-lapse as needed (say 10 minutes).

           

          But of course, this should be carefully tested

          on a non-production Tableau Server instance

          before even thinking about "go-to-production".

           

          Hope this could help a bit.

           

          Yours,

          Yuri

          • 2. Re: Publishing extracts is slow (performance issues)
            Jeff Strauss

            How do I check disk configuration?  I believe we have RAID-10, but I do not know if it's SAS-6 or SAS-12 or something else nor do I even know what this means, can you help?

            • 3. Re: Publishing extracts is slow (performance issues)
              Yuriy Fal

              Hi Jeffrey,

               

              These are questions to ask your HW systems engineer.

              Please ask about your physical server(s) vendor model,

              general HW config/ RAID config (if any) / HDD model(s).

              SAS-6 or SAS-12 meant Serial Attached SCSI interface (6Gbit/s or 12Gbit/s).

               

              If you have RDP access to your TS node(s), you may want

              to look at Windows Storage Manager and Device Manager

              for info about storage configuration & disk drives / volumes.

               

              It doesn't tell much if you're running TS on a virtual machine, though.

               

              Yours,

              Yuri

              1 of 1 people found this helpful
              • 4. Re: Publishing extracts is slow (performance issues)
                Jon Cantrell

                Thank you for the suggestions. I have passed them on to my server admin to see if he can implement some or all of the changes to improve performance.

                 

                -Jon

                • 5. Re: Publishing extracts is slow (performance issues)
                  Jeff Strauss

                  Our boxes are bare metal.  If you see any optimization opportunity, let me know.  I think we are good, but it never hurts to have another pair of eyes...

                   

                  - The data sector is RAID-10

                   

                  - Each span is SAS

                   

                  - Negotiate and capable speed is 6Gbps

                   

                   

                   

                  • 6. Re: Publishing extracts is slow (performance issues)
                    Yuriy Fal

                    Hi Jeffrey,

                     

                    I don't see anything to optimize further here.

                     

                    Yours,

                    Yuri

                     

                    PS  The question that thrills me here in this topic

                    is about which part of the whole publish transaction

                    is a bottleneck -- whether it an upload or a publish itself.

                    • 7. Re: Publishing extracts is slow (performance issues)
                      John Kuo

                      I wonder if your VM (I'm assuming you are using VM's) is tuned correctly. Check out attached document.

                       

                      Best,

                       

                      John

                      • 8. Re: Publishing extracts is slow (performance issues)
                        Eric Axelrod

                        When I see this behavior it's almost always related to poor network

                        performance between Tableau Desktop & Server.

                         

                        Are Desktop & Server on the same local network in the same datacenter?  Or

                        are you using Desktop on your local machine and publishing to a remote

                        server?

                         

                        The best test to isolate network issues is to (temporally) install Tableau

                        Desktop on the same machine as the server and retest.  If that results in a

                        huge performance variance then it's probably the upload speed of your

                        network.

                         

                        Eric Axelrod

                        President & Chief Architect

                        DIGR

                        www.digr.io