* Checkpoints commit all dirty pages to disk. Therefore the very last checkpoint is the true starting point for SQL Server to start recovery from, either by a restart or restore (crash of some type).
* SQL Server can perform the restart recovery on multiple databases at the same time, dependent on the number of threads it can spin up.
* ROLLBACKx are often MUCH slower that the actual transaction due to the fact that the transaction likely is multi-threaded and ROLLBACKS are mostly single-threaded.
* KILLing an open transaction causes SQL Server to roll it back
Views – 1839