Tableau Performance Optimization¶

Performance tuning in Tableau spans four layers: server load, data source, calculations, and view-level. The target is initial load < 10 seconds and filter switch < 6 seconds. Use the built-in Performance Recorder to identify bottlenecks.

Key Facts¶

Target performance: primary load < 10 sec, filter change < 6 sec
Performance Recorder: Help -> Settings and Performance -> Start Performance Recording
Server: add ?:record_performance=yes to URL
Report quality formula: (Informative content x Credibility x Engagement) / Load time
Peak load is non-linear - one heavy report can consume 75% of server CPU
Solution: optimize individual reports, not just scale the server

Patterns¶

Performance Recording Events¶

Event	Description
Compiling query	Generating SQL based on filters/calculations
Executing query	Sending query to DB or Extract
Computing layouts	Computing row-level calculated fields
Blending	Data blending operations
Geocoding	Placing points on map
Table Calculations	Computing table calculations
Rendering	Visual rendering

Server Load Optimizations¶

Fix dashboard sizes for frequently used dashboards (enables caching)
Delete unused reports, Extracts, and their scheduled refreshes
Reduce Extract size: Hide unused fields -> recreate Extract
Remove stale books from scheduled refreshes
No duplicate books or data sources
Group individual access permissions (reduces permission checking overhead)

Data Source Optimizations¶

Extract > Live in almost all cases (materialized data is faster)
Avoid Custom SQL in LIVE connections
Avoid GROUP BY and ORDER BY in Custom SQL
Use INNER JOIN with Referential Integrity flag enabled
Reduce row count: aggregate if not visualizing finest granularity
Non-additive data: store pre-aggregated by scale/slice rather than raw rows
For cross-granularity data: use Relations over Data Blending (or pre-join in DWH)

Calculation Optimizations¶

Prefer type hierarchy (faster to slower): Boolean > Numeric > Date > String
Use STARTSWITH over CONTAINS for string matching
Avoid LOD in complex nested calculations (each adds JOINs)
Materialize calculated fields in Extract before publishing
Move complex calculations to data source (SQL/ETL) when possible

View-Level Optimizations¶

Fewer sheets per workbook (>6 -> consider reducing)
Unique sheets over similar ones (duplicates = redundant queries)
Each worksheet generates one query to the source
Minimize marks on a single view (thousands of points = slow rendering)

Published vs Embedded Extract¶

Published Extract: shared across workbooks, but loses cache when used with relevant filters
Embedded Extract: embedded in workbook, better cache retention for that specific workbook

Data Blending Performance¶

Data Blending is inherently slow - avoid it. It creates multiple queries. Prefer: 1. Relations (Tableau 2020+) 2. Pre-prepared JOINs in the data warehouse 3. Only use Blending when truly independent sources need per-worksheet flexibility

Gotchas¶

Dashboard sizing affects caching: auto-sized dashboards can't be cached as effectively as fixed-size ones
Hiding unused fields and recreating Extract is a quick win that's often overlooked
Performance Recorder output shows time per event - focus on the longest bar first
Extract optimization (filter rows and columns at extract time) reduces file size but is a manual step that's easy to forget
Custom SQL in Live mode bypasses Tableau's query optimizer entirely