1 Reply Latest reply on May 21, 2012 8:01 AM by Russell Christopher

    data server hardware different from other workers

    Ben Sullins

      Greetings!

       

      I'm building out a Tableau Server environment in which I am separating out the Active Data Engine and Repository from all other server components. The challenge I have is that I want to scale this data engine environment to handle much larger capacity than the rest of the 'workers' running vizql and app server processes.

       

      Everything I've read however has said to keep the hardware the exact same across all nodes in the cluster. Does someone out there know how/if this is actually the case or if I CAN have the data engine run on a different hardware configuration?

       

      Thanks in advance!

       

      Cheers,

      Ben

        • 1. Re: data server hardware different from other workers
          Russell Christopher

          Hey Ben -

           

          You absolutely can drop the data engine and repository on a machine that is "beefier" than other workers in your farm.

           

          This could potentially become a "bad" thing if you're ALSO implementing the new High Availability features of Tableau Server 7.0.4 which allow the Repository and Data Engine to fail over to a different node. If this occurred and the node they fail over to is puny by comparison, you'll see performance degradation. Not a big surprise there though, right?

           

          Another thing that you'll want to consider - Backgrounders. Beginning in 7.0.2, these guys actually CREATE/WRITE extracts that are being refreshed, so they can suck a whole lot more resources (generally CPU) than they used to. This is actually good for the overall scalability of Server, but can catch folks unawares. To take this into account, latest / greatest best practices say to locate the data engine and backgrounders on different hardware than your vizqls and webapps - you wouldn't want a bunch of backgrounders which start refreshing extracts to suck up all your CPU and leave your poor vizqls with no resources to actually render reports

           

          Good luck!