Identity columns in the Cluster Edition behave differently from those in nonclustered editions of Adaptive Server. Although the Cluster Edition guarantees that identity values are unique, for performance reasons the values may not monotonically increase.
In nonclustered Adaptive Server, a set of identity values are burned into memory to reduce disk I/Os as inserts access the next value from memory. In the Cluster Edition, the same size set is burned into memory, but the set is split among the cluster instances. In a two-instance cluster with an identity set size of 250000, the first instance inserts values {1,2,3, and so on}, and the second instance inserts values {125000,125001,125002, and so on}.
The next_identity function reports the next identity value for a table from the instance in which next_identity is executed. For example, next_identity returns 4 for instance 1 and 125003 for instance 2.
The behavior of the identity_burn_max remains the same as for a nonclustered Adaptive Server because the burn size and burn behavior is unchanged in the Cluster Edition.
