Beginning Microsoft SQL Server 2008 ... - S3 Tech Training

Chapter 9: SQL Server Storage and Index Structures
The way the physical data is stored varies between clustered and non-clustered indexes. The way SQL
Server traverses the B-Tree to get to the end data varies between all three index types.
All SQL Server indexes have leaf level and non-leaf level pages. As we mentioned when we discussed B-
Trees, the leaf level is the level that holds the “key” to identifying the record, and the non-leaf level
pages are guides to the leaf level.
The indexes are built over either a clustered table (if the table has a clustered index) or what is called a
heap (what’s used for a table without a clustered index).
Clustered Tables
A clustered table is any table that has a clustered index on it. Clustered indexes are discussed in detail
shortly, but what they mean to the table is that the data is physically stored in a designated order. Individual
rows are uniquely identified through the use of the cluster key — the columns that define the clustered
index.
This should bring to mind the question, “What if the clustered index is not unique?” That is, how can a
clustered index be used to uniquely identify a row if the index is not a unique index? The answer lies
under the covers: SQL Server forces any clustered indexes to be unique — even if you don’t define them
that way. Fortunately, it does this in a way that doesn’t change how you use the index. You can still
insert duplicate rows if you wish, but SQL Server will add a suffix to the key internally to ensure that
the row has a unique identifier.
Heaps
A heap is any table that does not have a clustered index on it. In this case, a unique identifier, or row ID
(RID), is created based on a combination of the extent, pages, and row offset (places from the top of the
page) for that row. A RID is only necessary if there is no cluster key available (no clustered index).
Clustered Indexes
270
A clustered index is unique for any given table — you can only have one per table. You don’t have to have
a clustered index, but you’ll find it to be one of the most commonly chosen types for a variety of reasons
that will become apparent as we look at our index types.
What makes a clustered index special is that the leaf level of a clustered index is the actual data — that
is, the data is re-sorted to be stored in the same physical order of the index sort criteria state. This means
that once you get to the leaf level of the index, you’re done; you’re at the data. Any new record is inserted
according to its correct physical order in the clustered index. How new pages are created changes depending
on where the record needs to be inserted.
In the case of a new record that needs to be inserted into the middle of the index structure, a normal
page split occurs. The last half of the records from the old page are moved to the new page, and the new
record is inserted into the new or old page as appropriate.
In the case of a new record that is logically at the end of the index structure, a new page is created, but
only the new record is added to the new page, as shown in Figure 9-4.