dax-optimisation

Capture a server timing trace of the query, establish a warm-cache baseline, then apply matching Tier 1 patterns such as replacing FILTER table arguments with column predicates or caching repeated expressions in variables. Validate semantic equivalence after each change.

Trace capture options include the Tabular Editor CLI te query command, DAX Studio Server Timings, the TOM Trace API via PowerShell, Fabric Workspace Monitoring KQL queries, and the Power BI Modeling MCP. Each provides FE/SE duration splits and xmSQL events.

The formula engine handles all DAX logic single-threaded and is the bottleneck in most slow queries. The storage engine reads compressed VertiPaq data multi-threaded but supports only basic aggregations and joins. Optimization means pushing work to the SE and eliminating callbacks.

Few SE queries with high duration and parallelism near 1.0 indicate a data layout problem, not a DAX problem. Check segment sizing, column cardinality, and compression; for Direct Lake, apply V-Ordering and optimize Parquet rowgroup sizes.

Tier 1 rewrites preserve semantics by modifying only measure definitions while keeping EVALUATE and grouping identical. Tier 2 and above changes alter output shape or model structure, so they require explicit user approval before applying.

No. Dynamic format strings evaluate once per rendered cell and never appear in query traces. A format string referencing its own measure can roughly double visual render time; test by swapping to a static format string and re-timing the visual.