SQLCAT Technical Note: Resolving PageLatch Contention on Highly Concurrent Insert Workloads Part 1

The article states that they couldn't add a leading column to an index because it would break the application. This is strange because indexes have nothing to do with the logical model & can't possibly break applications unless they change constraints. Even stranger, the solution published goes on to actually change the logical design by adding an additional column! I found this confused the message they were trying to get accross..

Perhaps they meant to say "constraint" instead of "index" but even this would be irrelevant as you'd simply move the constraint to an identical (unique) index & design the cluster key as described with the non sequential leading key.

The real problem with changing the leading key of the clustered index is that it vastly increases the incidence of page splits vs the hash key solution published, as insertion into each quarter of an index (with a single partition) will obviously experience many more splits than at the end of each partition's allocation.

Last, the solution published could have been improved further by hiding the partitioned table under a view that doesn't include the new column to reduce the likelihood of breaking the application.