|
|
Posted by Alexander Kuznetsov on 04/18/06 00:16
Erland Sommarskog wrote:
> That explains why I haven't heard of it. SQL Server performs no such
> deliberations, as far as I know.
Well if on the leaf level of a non clustered index you've got an RID, a
structure looking like (extent#, page#, row_on_page#), and you know
that matching rows are likely to be stored on adjacent pages, you can
traverse a range of the index, sort RIDs of matching rows, and read
every page with matching rows only once. This is why clustering factor
is a very useful piece of statistics for DB2/Oracle.
If on the leaf level of a non clustered index you've got a bookmark,
then you just know absolutely nothing about the physical location of
the rows you need to retrieve. I don't see any way SQL Server could
utilize clustering factor as long as bookmarks are used to access rows
from non-clustered indexes (unless it's a heap table, which is not the
best practice). Please correct me if I'm wrong.
Navigation:
[Reply to this message]
|