Wednesday 4 January 2012

Indexed View

Views allow you to create a virtual table by defining a query against one or more tables. With a standard view, the result is not stored in the database. Instead, the result set is determined at the time a query utilizing that view is executed.

Creating a unique clustered index on a view changes it to an indexed view. The clustered index is stored in SQL Server and updated like any other clustered index, providing SQL Server with another place to look to potentially optimize a query utilizing the indexed view.

Queries that don't specifically use the indexed view can even benefit from the existence of the clustered index from the view. In the developer and enterprise editions of SQL Server, the optimizer can use the indexes of views to optimize queries that do not specify the indexed view. In the other editions of SQL Server, however, the query must include the indexed view and specify the hint NOEXPAND to get the benefit of the index on the view.

If your queries could benefit from having more than one index on the view, non-clustered indexes can also be created on the view. This would supply the optimizer with more possibilities to speed up the queries referencing the columns included in the view.

Where to Use Them

Indexed views have both a benefit and a cost. The cost of an indexed view is on the maintenance of the clustered index (and any non-clustered indexes you may choose to add). One must weigh the cost to maintain the index against the benefit of query optimization provided by the index. When the underlying tables are subject to significant inserts, updates, and deletes, be very careful in selecting the indexes (both table and view) that will provide the greatest coverage across your queries for the lowest cost.

Typically, environments that are best suited for indexed views are data warehouses, data marts, OLAP databases, and the like. Transactional environments are less suitable for indexed views. Look for repeating joins utilizing the same columns, joins on large tables, aggregations on large tables, and repeating queries as potential candidates for indexed views. Be careful of creating indexed views where the result set contains more rows than the base tables as this will be counterproductive.

How to Create Them

A view that is to be indexed has to be created with schema binding. This means that once the indexed view is created, the underlying tables cannot be altered in any way that would materially affect the indexed view unless the view is first altered or dropped. It also means that all the tables referenced in the view must be referenced by their two-part name (schemaname.tablename).

Below is an example of the CREATE statement for an indexed view, MyView, and its underlying table, MyBigTable. The table is first created, then the view that references two of the table's three columns, and finally the unique clustered index on the view making it an indexed view.

CREATE TABLE MyBigTable(

ItemID            INT PRIMARY KEY,

ItemDsc           VARCHAR(20),

QTY               INT)

GO

 

CREATE VIEW MyView WITH SCHEMABINDING AS

SELECT ItemID, QTY

  FROM dbo.MyBigTable

 WHERE QTY > 10

GO

CREATE UNIQUE CLUSTERED INDEX idx_MyView ON MyView(QTY)

Once this index is created, the result set of this view is stored in the database just like any other clustered index. Any query that explicitly uses the view will be able to take advantage of the index on the view. Queries that contain a predicate similar to the view and that fall into the range defined by the view may also reap the optimization rewards of having that index available (assuming the execution cost is non-trivial). Consider the following query:

SELECT ItemID

  FROM MyBigTable

 WHERE QTY > 30

Even though the query does not use the indexed view, the optimizer has the option of using the clustered index created on the view if it provided better performance than the clustered or non-clustered indexes on the base table.

If you want the optimizer to always choose the indexed view over the base tables when optimizing a query containing an index view, you must use the hint NOEXPAND. Conversely, if you'd like to see how a query containing an indexed view would perform utilizing the base tables instead, you can specify the option EXPAND VIEWS, thus saving you the time substituting the base tables yourself.

Constraints

An index cannot be created on just any view. Several constraints exist that a view must meet in order for the index creation to be successful. We discussed WITH SCHEMABINDING and two-part table names above. Here are some other constraints.

  • The view must have been created with certain SET options, such as QUOTED_IDENTIFIER and CONCAT_NULL_YIELDS_NULL set to ON.
  • The session creating the index must also have the correct SET options.
  • Any user-defined functions referenced by the view must have been created using WITH SCHEMABINDING.
  • The view must be deterministic (consistently providing the same result given the same input).
  • The base tables must have been created with the proper ANSI_NULLS setting.
  • The result set of the view is physically stored in the database, thus storage space for the clustered index is also a constraint to consider.

In addition to this, there are constraints on the contents of the view. For instance, the view may not contain EXISTS or NOT EXISTS, OUTER JOIN, COUNT(*), MIN, MAX, sub queries, table hints, TOP, UNION, and much more. Check the SQL Server Development Center on MSDN for a complete listing.

Conclusion

Indexed Views are a great way to realize performance gains under the right circumstances. Be aware of the costs of creating indexed views in highly transactional environments. If your environment happens to be one of more querying than updating, however, indexed views might be just what the optimizer ordered.

 

Posted By: MR. JOYDEEP DAS

 

No comments:

Post a Comment