Trying to better understand row-versioning based transaction isolations - RCSI (Read Committed Snapshot Isolation and SNAPSHOT) - in SQL Server...
MSDN online article "Choosing Row Versioning-based Isolation Levels" states:
- "For most applications, read committed isolation using row versioning is recommended over snapshot isolation for the following reasons:
- It consumes less tempdb space than snapshot isolation"
which contradicts to the statement from Kimberly L. Tripp, Neal Graves's "SQL Server 2005 Row Versioning-Based Transaction Isolation":
- "Row Versioning in Read Committed Using Row Versioning
For queries that are running under read committed using row versioning, row versions are necessary only for the duration of each SELECT statement in the transaction. However, row versions are maintained until the end of the transaction. This is an important point to consider for tempdb storage. The storage impact will be the same as it would be for snapshot isolation."
So, why does RCSI use less (or same) space?
Probably I formulated the original question bad but it implies that there are RTFM explanations and confirmations for both contradictory "theories" and I'd like to understand:
Which is correct?
Why is the one is really correct?
Why isn't another really correct (Why is another wrong?)?