The COMPRESSION_DELAY option keeps a closed delta rowgroup uncompressed for a number of minutes, so rows that change soon after they arrive are cheaper to modify. This post loads two demo tables, updates fresh rows in each, and shows the deleted rows that early compression leaves behind.
Fragmentation in Columnstore Indexes: Find and Fix It
Fragmentation in columnstore indexes means deleted rows that stay inside compressed row groups, and row groups that are too small. This post builds a demo table, measures the deleted rows with the row group view, and compares REORGANIZE with REBUILD.
Table Usage in the Plan Cache: Count Queries That Touch a Table
Table usage in the plan cache shows how many times the cached queries of a database touched each table. This post builds a demo database, shows why a text search of the cache misleads, reads the table names from the plan XML, and lists the queries behind one table.
Disable Row Goal in SQL Server: When the Hint Helps and When It Hurts
To disable row goal for one query, add the DISABLE_OPTIMIZER_ROWGOAL hint to it. This post builds a table where the row goal chooses a bad plan, shows the reads with and without the hint, and then shows a table where the row goal wins.
Number of Rows Read in an Execution Plan: What It Tells You
The number of rows read shows how many rows a scan or seek touched, and it can be far larger than the rows it returned. This post builds a demo table, compares a scan with two seeks, and explains why a seek can still read too many rows.
Instance Level Fill Factor or Index Level: What to Set
An instance level fill factor is a default for every index that sets none, and an index level setting wins. This post reads the setting, measures what each fill factor costs in pages and saves in page splits, and gives a plan for choosing the level.
Slowest Cached Queries in SQL Server: Time, Reads and Plan
To find the slowest cached queries in SQL Server, rank each statement by its average elapsed time and keep its plan beside it. This post builds a small workload, reads sys.dm_exec_query_stats with the plan, and shows why a database filter on the sql text hides every ad hoc query.







