Indexes: An Overview and Maintenance for Performance
Many people know the importance of creating indexes on SQL Server database tables. Indexes greatly improve the performance of a database. However, while many people create indexes on their SQL Server tables, many people don’t maintain them properly to ensure queries run efficiently as possible. I’ll begin by giving a quick overview of how SQL Server 2000 stores data and how indexes improve performance. Then, I’ll spend quite a bit of time explaining why, when, and how to maintain indexes with DBCC SHOWCONTIG and DBCC INDEXDEFRAG to ensure queries run in the most efficient manner.
SQL Server 2000 stores data into what is known as a heap. A heap is a collection of data pages containing rows for a table. The data isn’t stored in any particular order and the data pages themselves aren’t in any sequential order. The data is just there with no real form or organization. When SQL Server accesses data in this form, it does a table scan. This means SQL Server starts reading at the beginning of the table and scans every page until it finds the data that meets the criteria of the query. If a table is very large, this could greatly decrease the performance of queries.
Indexes will hasten the retrieval of data. It is important to understand how data is used, the types of queries being performed and the frequency of the queries that are typically performed when planning to create indexes. An index is far more efficient when the query results return a low percentage of rows and the selectivity is high. High selectivity means a query is written so it returns the lowest number of rows possible. As a rule, indexes should be created on columns that are commonly searched; this includes primary and foreign keys. It follows that columns that contain few unique values should never be indexed; this will increase the number of rows returned in a query.
There are two types of indexes to consider when planning: Non-Clustered and Clustered Indexes.
A non-clustered index stores data comparable to the index of a text book. The index is created in a different location than the actual data. The structure creates an index with a pointer that points to the actual location of the data. Non-clustered indexes should be created on columns where the selectivity of query ranges from highly selective to unique. These indexes are useful when providing multiple ways to search data is desired.
A clustered index stores data similar to a phone directory where all people with the same last name are grouped together. SQL Server will quickly search a table with a clustered index while the index itself determines the sequence in which rows are stored in a table. Clustered indexes are useful for columns searched frequently for ranges of values, or are accessed in sorted order.
While I could go on and on about how SQL Server 2000 stores and accesses data in a heap and in an Index architecture, I will move on to discuss maintaining indexes with DBCC SHOWCONTIG and DBCC INDEXDEFRAG.
Once indexes have been created, it is important to maintain indexes to ensure the best possible performance. If indexes are not maintained, over time the data will become fragmented. Fragmentation is the inefficient use of pages within an index*. There are a number of tools available that will help with optimizing indexes to ensure they are running well, however I will only discuss DBCC SHOWCONTIG and DBCC INDEXDEFRAG in this article.
The DBCC SHOWCONTIG command will provide fragmentation information on data and indexes within a specified table and it will also determine if the data and index pages are full. If a page is full, SQL Server must split the page to make room for new rows. This statement should be run on heavily modified tables, tables that contain imported data, or tables that seem to cause poor query performance. When the statement is executed, here is what will be returned:
Statistic Description
Pages Scanned Number of pages in the table or index.
Extents Scanned Number of extents in the table or index.
Extent Switches Number of times the DBCC statement moved from one extent to another while it traversed the pages of the table or index.
Avg. Pages per Extent Number of pages per extent in the page chain.
Scan Density [Best Count: Actual Count] Best count is the ideal number of extent changes if everything is contiguously linked. Actual count is the actual number of extent changes. The number in scan density is 100 if everything is contiguous; if it is less than 100, some fragmentation exists. Scan density is a percentage.
Logical Scan Fragmentation Percentage of out-of-order pages returned from scanning the leaf pages of an index. This number is not relevant to heaps and text indexes. An out of order page is one for which the next page indicated in an IAM is a different page than the page pointed to by the next page pointer in the leaf page.
Extent Scan Fragmentation Percentage of out-of-order extents in scanning the leaf pages of an index. This number is not relevant to heaps. An out-of-order extent is one for which the extent containing the current page for an index is not physically the next extent after the extent containing the previous page for an index.