Distributed Installation Recommendations

When you add nodes to a Tableau Server installation, you must decide how many processes to run on each computer. This page provides some general recommendations that are intended only as a starting point.

In addition to these general recommendations, you should also:

  • Understand how your organisation uses Tableau Server and tune your configuration for your use case – for example, whether you want to optimise for user response or for extract refreshes.

  • Perform thorough performance testing to identify the best places to adjust process configuration.

For more information on tailoring a Tableau Server installation to your organisation's needs, see Performance Tuning Examples.

For more information on the requirements for a distributed installation and for information on configuring additional nodes, see Distributed and High Availability Tableau Server Installations.

Recommendations for all installations

Although the computers that make up a Tableau Server cluster do not need to have identical hardware, they must all meet the same minimum system requirements. The recommendations on this page assume that the computers where you install Tableau Server have eight cores or more.

The following recommendations apply to all server configurations:

  • Run Backgrounder processes on a dedicated computer if you plan on refreshing extracts frequently. Backgrounder processes are generally the most CPU intensive and can slow down other processes on the same computer.

    Note: In versions earlier than 10.5, the backgrounder process spun up temporary instances of Data Engine when refreshing extracts. In 10.5, a single instance of Data Engine is installed with the backgrounder during setup. The backgrounder uses this running instance of Data Engine when refreshing extracts.

  • If you plan to refresh extracts frequently or if you plan to refresh large extracts, increase the number of processes for Backgrounder processes.

  • Run VizQL processes on a different computer than Backgrounder processes. Having them on the same machine means that extract refreshes can affect user views.

  • The instance of Data Engine installed on the node where File Store is installed is used for querying data for view requests. Consider separating the File Store process from the backgrounder processes to help minimise the backgrounder tasks from affecting user views.

  • Optimising with topology configurations:

    • Co-locating File Store on the same node as the Administration Controller can reduce the length of time it takes to back up Tableau Server by reducing or eliminating the need to transfer data between nodes during the backup process. This is especially true if your organisation uses many extracts.
    • Co-locating the repository (pgsql) with the Administration Controller node can also help to reduce back up time, but the time savings is less significant than that of the File Store.

    The Administration Controller is usually on the initial node, unless you have had an initial node failure and moved the controller to another node.

Note: In a distributed installation with three or more nodes, you can have a maximum of two repository instances (active and passive). You can also run Tableau Server with one repository, but doing this means there is no failover available for the repository. For more information, see Tableau Server Repository.

Thanks for your feedback!