SQL Server heap v.s. clustered index

Heap storage has nothing to do with these heaps.

Heap just means records themselves are not ordered (i. e. not linked to one another).

When you insert a record, it just gets inserted into the free space the database finds.

Updating a row in a heap based table does not affect other records (though it affects secondary indexes)

If you create a secondary index on a HEAP table, the RID (a kind of a physical pointer to the storage space) is used as a row pointer.

Clustered index means that the records are part of a B-Tree. When you insert a record, the B-Tree needs to be relinked.

Updating a row in a clustered table causes relinking of the B-Tree, i. e. updating internal pointers in other records.

If you create a secondary index on a clustered table, the value of the clustered index key is used as a row pointer.

This means a clustered index should be unique. If a clustered index is not unique, a special hidden column called uniquifier is appended to the index key that makes if unique (and larger in size).

It is also worth noting that creating a secondary index on a column makes the values or the clustered index's key to be the part of the secondayry index's key.

By creating an index on a clustered table, you in fact always get a composite index

CREATE UNIQUE CLUSTERED INDEX CX_mytable_1234 (col1, col2, col3, col4)CREATE INDEX IX_mytable_5678 (col5, col6, col7, col8)

Index IX_mytable_5678 is in fact an index on the following columns:

col5col6col7col8col1col2col3col4

This has one more side effect:

A DESC condition in a single-column index on a clustered table makes sense in SQL Server

This index:

CREATE INDEX IX_mytable ON mytable (col1)

can be used in a query like this:

SELECT  TOP 100 *FROM    mytableORDER BY       col1, id

, while this one:

CREATE INDEX IX_mytable ON mytable (col1 DESC)

can be used in a query like this:

SELECT  TOP 100 *FROM    mytableORDER BY       col1, id DESC

Heaps are just tables without a clustering key - without a key that enforces a certain physical order.

I would not really recommend having heaps at any time - except maybe if you use a table temporarily to bulk-load an external file, and then distribute those rows to other tables.

In every other case, I would strongly recommend using a clustering key. SQL Server will use the Primary Key as the clustering key by default - which is a good choice, in most cases. UNLESS you use a GUID (UNIQUEIDENTIFIER) as your primary key, in which case using that as your clustering key is a horrible idea.

See Kimberly Tripp's excellent blog posts GUIDs as Primary and/or the clustering key and The Clustered Index Debate Continues for excellent explanations why you should always have a clustering key, and why a GUID is a horrible clustering key.

My recommendation would be:

• in 99% of all cases try to use a INT IDENTITY as your primary key and let SQL Server make that the clustering key as well
• exception #1: if you're bulk loading huge data amounts, you might be fine without a primary / clustering key for your temporary table
• exception #2: if you must use a GUID as your primary key, then set your clustering key to a different column - preferably a INT IDENTITY - and I would even create a separate INT column just for that purpose, if no other column can be used

Marc

Books Online is the best source!

The whole Database Engine - Planning and Architecture - Tables and Index Data Structures Architecture is very good internal introduction.

From this link you can download a local copy of Books Online(it is free). It is the best (and official) reference to all Sql 2008 questions.