Let me show you why data types still matter
summary
Shows concrete failures and fixes: float loses precision so casting and ceiling produced wrong truth values and ruined a cursor-to-set refactor; use float only for imprecise measurements not counts or money. Demonstrates concatenation and type precedence where mixing varchar and nvarchar silently truncates strings so nvarchar halves capacity unless both sides use MAX. Fixing datetime2 precision enables partition elimination and cuts an ETL from 12 hours to under 2. Also cites memory grant bloat (queries asking ~80 GB) and costly implicit conversions causing index scans.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:52Float data type surprises
- 4:45Dangers of float for precision
- 5:07String concatenation issues
- 7:39Unicode vs non-Unicode strings
- 8:24Choosing string column types
- 10:13Data types and performance
- 12:00Memory grants and data types
- 12:20Partition elimination precision
- 15:34Implicit conversion costs
- 18:00Questions and closing remarks
related talks
19mLet me show you why data types still matter- 19mLet me show you why data types still matter
46mExacting Standards: Considerations for Precision and Accuracy ...
46mDon't do that!
19mNO ARRAY OF SUNSHINE: an introduction to data types from a data quality perspective
54mLessons learned from refactoring multi terabyte databases for the last 10 years
52mWriting T-SQL code for the engine, not for you "more demos" by popular demand
53mSQL Server Internals: The Practical Angle