Conor vs. Row Growth in Read Committed Snapshot Isolation
At PASS last week, I was asked about fragmentation caused by RCSI when a row is first converted to use RCSI.
I talked to some people in the team and found that one of my colleagues had already written a nice blog entry about it.
Basically, the overhead is taken once and you can do index rebuild to complete the process if desired.
RCSI is still a pretty cool feature and I think people should consider using it when they have the opportunity – it takes awhile to get ones head wrapped around the concept of optimistic concurrency/versioning, but it can be very effective at reducing contention in applications.