When Adaptive Server splits an index or data page, it moves some rows from the original pages to the newly created page. The operation of moving the rows is not logged. Adaptive Server 15.0.2 uses asynchronous writes to ensure consistency.
Adaptive Server writes the new page to disk before writing the modified page (with the rows removed) to disk. This ensures that Adaptive Server can restore the previous version of the page if the transaction is undone. Adaptive Server can find these rows on the new page and move them back to the old page even if the rows are missing in the old page, and their row contents were not logged.
The new page reaches the disk before the transaction commits, which ensures that Adaptive Server cannot loose the committed data. If the transaction was committed, Adaptive Server is not required to redo the transaction for the new page, which would be impossible since the movement of the rows was not logged. In the case of undo, the new page’s allocation is backed out; there's no page pre-image to restore.
Previous versions of Adaptive Server ensured these two conditions were met by synchronously writing the new page to disk. However, because the server could block until the synchronous write returned, this caused a degradation in performance.
Adaptive Server version 15.0.2 uses asynchronous writes to disk that satisfy the conditions described above and do not require the server to block as it waits for the write to complete.
Adaptive Server version 15.0.2 uses these asynchronous writes automatically and requires no configuration on your part.