5 Replies Latest reply on May 22, 2018 2:01 PM by Mark McGhee

    Server specifications and configuration

    John Hobby

      Howdy fellow SA's,

       

      So, this is more of a discussion rather than a full blown question.  My environment is spec'ed out as the following.

       

      Server version:  10.4

      Hardware:  4 VM's each with win 2012 r2 64bit / 8 cores / 32g memory / 500g hard drive

      Layout:  1 primary / 1 backup primary / 2 workers

      Users:  ~250 seat licenses

      Extracts: 1500+ (expected to grow to close to 2000 by the end of 2018)

       

      Right now, I have two worker machines (with the above specs) that do most of the processes.  They are also configured to run 8 backgrounders each, meaning I can refresh 16 total extracts at one time.  This has been working successfully so far.  We have around 1500 extracts that have to be refreshed over the weekend.  I did have to adjust some schedules to keep some errors from happening, but those have gone away.  We also built some staging table for historical data that doesn't change, so the script to get the data runs a little faster.

       

      How are other servers configured?  Anyone have machine dedicated to just backgrounders and all other processes on different machines?  If so, what are the machine specs?  How many of each processes do ya'll have setup for each machine?  How's performance?

       

      Below is what my server looks like based on the processes assigned.

       

      Thanks,

      John

        • 1. Re: Server specifications and configuration
          Mark McGhee

          Hi John,

           

          From a support perspective there are a few items that thought worth mentioning.   But if you're system has been performing well these may not be as applicable to your situation:

           

          (1)  Memory - For Production systems we typically like to see 8GB of memory per core.  For both workers there's a chance to increase performance by bumping up the memory to 64GB each.

          (2)  Backgrounder processes can be very resource hungry.  If you have a "dedicated backgrounder" then we recommend you can go to a 1:1 ratio of 1 Backgrounder per core.  But if there are other Tableau processes on a node such as VizQL then we tend to have (# of cores/2).   Part of the reason is that VizQL can be resource intensive as well and they could get into a scenario where they're competing for resources.  The same could be said if too many backgrounders are placed on one node as well (though totally understand the temptation...  :-)

           

          Not sure if you've tried it out be we do have a LogShark tool that can parse through Tableau Server log files and it creates workbooks for essentially each component.  One is for Backgrounder which would give you an idea just how busy the backgrounders have been.  Another measures CPU/memory and it will reveal if Tableau Server has had to kill off processes (this would be the ResourceManager workbook).  Below is a link that provides more information about the tool:

           

          Introducing Logshark: Analyze your Tableau Server log files with Tableau | Tableau Software

           

          One thing worth noting, LogShark is *not* a realtime log monitoring tool so there's no advantage of installing it on any node in a Tableau cluster.  In fact would recommend not installing it there but it doesn't have to be a powerful machine to run it.  I actually run the tool on my laptop.

          • 2. Re: Server specifications and configuration
            Toby Erkson

             

            We pretty much have extracts running 24x7 serving primary users in Canada, USA, Mexico, & South America.  We are a global company so our Tableau Server is available to those over-seas but they typically use our TS as a POC and justification to get their own local one (which makes sense).

             

            Primary

            Tableau Server 10.5.2

            Core-based licensing

            16 cores

            196GB RAM

            1.99TB hard disk

            Windows Server 2008 R2 but am currently working with the VM team to upgrade it to Windows Server 2012 R2.

            Worker

            8 cores

            65GB RAM

            1.99TB hard disk

            Windows Server 2012 R2.

             

            I would like a second worker with 8 cores dedicated to Backgrounders since subscriptions are steadily increasing as well.

            • 3. Re: Server specifications and configuration
              John Hobby

              Thanks Mark - I'm hoping to present a case to our exec team to be able to beef up our systems.  I do want dedicated machines to just run the backgrounders (extract building).  Right now, they only gave me 32g mem but hoping to get this bumped to 64.  I'll look into LogShark and see about setting it up.  It does look like it requires MongoDB / PostgresSQL to be installed - so will read through the docs on those requirements.  With our extracts now pushing over 2500, I had to push the machines to their limits.  I ran into some weird errors, where an extract would get canceled about half way through - it seemed to be memory related.  Downloading a ton of records was eating up the memory, and when you have 16 extracts that can do that ... it figures it would error out.  I got around this by spreading out the start times of our extracts.  So, now on Sunday's I have about 800 extracts that run and instead of starting them all at 6am, I spread them across 6 different starting schedules and so far everything has been running pretty good.

               

              Hey Toby

              Curious, how many users do you have?  What about the number of data sources?  With your extracts running all the time, has anyone complained about the refresh time?   

               

              We noticed some performance issues extracting against Oracle - our financial extract has about 160 columns x (1m-30m) records, depending on the company.  What we did was to create a 'history' table for the financials that shouldn't change anymore (like years 2000 - 2017), then union'd that with selecting the 2018 records.  That cut our download time in half for a lot of extracts.

               

              How are your subscriptions running?  We haven't turned ours on yet, but that is something on our road-map over the next year.  Will you put your subscriptions on their own machine or you planning to have a few other processes running on it also?

               

              Thanks,

              John

              • 4. Re: Server specifications and configuration
                Toby Erkson

                30K registered Users but less than 3K have actually logged in.

                Just under 400 Data Sources.

                Data Source types:

                Excel, Access, Oracle, IBM DB2 & BLU, SAP HANA, Salesforce, SharePoint, .tde published by Alteryx, MS SQL Server, OData, Spatial, & of course PostgreSQL.

                Subscriptions (do it! ) are fine and it's my understanding that they run on the Backgrounders so there is no "subscriptions-only" worker needed, just more RAM and Backgrounders.

                 

                Our network is really the bottleneck, especially for those outside Oregon (our data center is in Oregon).  While we do have Live connections the majority are extracts in order to improve performance.

                • 5. Re: Server specifications and configuration
                  Mark McGhee

                  Hi John,

                   

                  Just a quick follow up.  You're correct in regards to LogShark but it already comes with the Mongo DB as part of the Setup program.  It does require a separate instance of Postgres being installed (you definitely do not want to try to loop results into any Tableau Server Postgres instance, they should be kept separate) but it doesn't take up as much space as you'd think.