Indexes in ms sql server
Download as DOCX, PDF0 likes199 views
Indexes in MS SQL Server allow for optimized data retrieval from tables. There are two main types of indexes - clustered and non-clustered. A clustered index physically orders the data based on the index keys. A non-clustered index logically orders the data and stores pointers to the physical data. The example shows how indexes can reduce the logical reads needed to retrieve data from 200,000 records in a table based on different query criteria. Fragmentation can occur internally within pages or externally across extents if data is stored non-contiguously on disk.
Indexes in ms sql server
- 1. Indexes in MS SQL Server Introduction I bought a book from crossword; he packed the book and added two bookmarks into my pack. A thought came to my mind. Why do I need a bookmark? I can easily memorize the page number and the next time resume from the same page when I resume reading, or read them all over to reach to the point where I stopped reading. But not all have a blessed memory; moreover, there are better things to remember, my grandpa would rather bookmark and rely on it to help him resume reading. It’s a kind of simple index, isn’t it? This article focuses on how MS SQL Server uses indexes to read and write data. Data is arranged by SQL Server in the form of extents and pages. Each extent is of size 64 KB, having 8 pages of 8KB sizes. An extent may have data from multiple or same table, but each page holds data from a single table only. Logically, data is stored in record sets in the table. We have fields (columns) identifying the type of data contained in each of the record sets. A table is nothing but a collection of record sets; by default, rows are stored in the form of heaps unless a clustered index has been defined on the table, in which case, record sets are sorted and stored on the clustered index. The heaps structure is a simple arrangement where the inserted record is stored in the next available space on the table page. Heaps seem a great option when the motive is simply storing data, but when data retrieval steps in, this option back fires. An index acts as a fire fighter in this scenario. Indexes are arranged in the form of a B-Tree where the leaf node holds the data or a pointer to the data. Since the stored data is in a sorted order, indexes precisely know which record is sitting where. Hence an index optimizes and enhances the data retrieval immensely. But everything comes at a cost; the price we pay for having an index on the table is, each time there is an Insert/Update/Delete, SQL Server updates the active indexes on the table where these DML are operated. Hence simply creating indexes madly for the sake of better data retrieval will not serve the purpose. If there are 20 indexes on a table, each time a DML is done on the table, all these 20 indexes shall be updated so that they can uniquely figure out the location of the record. Let’s dive deep into the indexes. Setup: All code has been tested on MS SQL Server 2008 R2. Clustered Index (CI) A clustered index is something that reorganizes the way records in the table are physically stored. Therefore a table can have only one clustered index. The leaf nodes of a clustered index contain the data pages, by which I mean the key-value pair in the clustered index has the index key and the actual data value. Also remember, a clustered index will be created on a table by default the moment a primary key is created on the table. A clustered index is something like your train ticket B4/24, you know that you need to board coach B4 and sit on seat number 24. So this index physically leads you to your actual seat.
- 2. We will follow this up closely with an example: Hide Copy Code USE TestDB GO CREATE TABLE Sales( ID INT IDENTITY(1,1) ,ProductCode VARCHAR(20) ,Price FLOAT(53) ,DateTransaction DATETIME); I have created a table Sales, and then created a Stored Procedure to insert 2,00,000 records into the Sales table. This sizable chunk of data will help us to notice the differences very clearly. Hide Copy Code CREATE PROCEDURE InsertIntoSales AS SET NOCOUNT ON BEGIN DECLARE @PC VARCHAR(20)='A12CB' DECLARE @Price INT = 50 DECLARE @COUNT INT = 0 WHILE @COUNT<200000 BEGIN SET @PC=@PC+CAST(@COUNT AS VARCHAR(20)) SET @Price=@Price+@COUNT INSERT INTO Sales VALUES (@PC,@Price,GETDATE()) SET @PC='A12CB' SET @Price=50 SET @COUNT+=1 END END EXEC InsertIntoSales Now we have created the table and inserted 2,00,000 records into it, but there is no index defined on any column. Press Control+M. This will “Include the Actual Execution Plan” in the results. Let's run the below query. Hide Copy Code
- 3. SET STATISTICS IO ON SELECT * FROM Sales WHERE ID=189923 ID ProductCode Price DateTransaction ----------- ---------------- ---------- ----------------------- 189923 A12CB189922 189972 2011-03-21 12:07:48.310 (1 row(s) affected) Table 'Sales'. Scan count 1, logical reads 1129, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0. (1 row(s) affected) The Execution plan tab on the results show that the record has been retrieved on a table scan and the logical reads are 1129. Now let’s build a clustered index on the ID column of the Sales table. Hide Copy Code CREATE CLUSTERED INDEX CL_ID ON SALES(ID); Let us press CTRL+M and rerun the same query: Hide Copy Code SET STATISTICS IO ON SELECT * FROM Sales WHERE ID=189923 ID ProductCode Price DateTransaction ----------- ---------------- ------------ ----------------------- 189923 A12CB189922 189972 2011-03-21 12:07:48.310
- 4. (1 row(s) affected) Table 'Sales'. Scan count 1, logical reads 3, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0. (1 row(s) affected) The Execution plan tab on the results shows that the record has been retrieved on Index seek and the logical reads are 3. After the clustered index creation, SQL Server has been able to reduce the logical reads dramatically and the query has been optimized. Clearly the index knows where to look for the record. Non-Clustered Index (NCI) A non-clustered index is a special type of index in which the logical order of the index does not match the physical stored order of the rows on disk. The leaf node of a non-clustered index does not consist of the data pages but a pointer to it. That goes to say that a non-clustered index can’t survive on its own - it needs a base to live on. A non-clustered index uses a clustered index (if defined) or the heap to build itself. When a non-clustered index uses the heap, the leaf node (or the pointer) is a physical location of the data. When it uses a clustered index, the leaf node (or the pointer) is the clustered index key value and this key value in turn points to the actual data. Part I: When NCI Uses a CI Getting back to Sales, we already have a CI (CL_ID) on the ID column, now if we have a query something like: Hide Copy Code SET STATISTICS IO ON SELECT * FROM Sales WHERE ProductCode like 'A12CB908%' order by Price Press Control+M and execute the query
- 5. There are arround 111 records retrived ----------------------------------------------------- (111 row(s) affected) Table 'Sales'. Scan count 1, logical reads 1130, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0. (1 row(s) affected) We find that the query first uses the clustered index to get 111 records and then uses a sort operation; the logical reads are as high as 1130. There is also a missing index suggestion. Let’s consider SQL Server’s advice and create a non-clustered index (NONCI_PC) on the ProductCode column. Since we have a CI already, this NCI would be built on the CI. Hide Copy Code CREATE NONCLUSTERED INDEX NONCI_PC ON SALES(ProductCode); Press Control+M and rerun the same query, this time around, we can see the data fetch plan change. Hide Copy Code SET STATISTICS IO ON SELECT * FROM Sales WHERE ProductCode like 'A12CB908%' order by Price ------------------------------- (111 row(s) affected) Table 'Sales'. Scan count 1, logical reads 351, physical reads 0, read-ahead reads 7, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0. (1 row(s) affected)
- 6. The logical reads have been minimized and the revised execution plan is as in the figure. This was the example where anon-clustered index used a clustered index. Part II: When NCI Uses a Heap When there is no clustered index built on a table and a non-clustered index is built, it uses the heap for data retrieval. The indexed column or columns are sorted along with a pointer to the physical location of the data. The big question is, how do I know if I should create an NCI on a CI or on a heap? The answer is in the query, if data is queried typically on one particular column, it would be beneficial to build NCI upon a CI, but when the case is something like what we have in the present Sales example where we will be building a one-column NC index on a heap, the NCI would be merely a two-column table containing the key-value pair (index key and physical location, i.e., value). This would be the best optimization in this scenario. To follow this up, let’s review the example closely. With respect to the sales example, let’s delete the clustered index CL_ID created on the ID column and re-evaluate. Hide Copy Code DROP INDEX Sales.CL_ID; SET STATISTICS IO ON SELECT * FROM Sales WHERE ProductCode like 'A12CB908%' order by Price ------------------------------------ (111 row(s) affected) Table 'Sales'. Scan count 1, logical reads 114, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
- 7. (1 row(s) affected) The logical reads have been further optimized and the execution plan also has been revised. In this case, the query uses the non-clustered index to be run on the heap. We have been able to create indexes so that our queries work with a minimal performance overhead. So now the next big question is, Wouldn’t it be great if there was someone to help us out in prompting on the indexes to be built based on our queries? FRAGMENTATION: Storing data non-contiguously on disk is known as fragmentation. Internal Fragmentation: When records are stored non-contiguously inside the page, then it is called internal fragmentation. External Fragmentation: When on disk, the physical storage of pages and extents is not contiguous. When the extents of a table are not physically stored contiguously on disk, switching from one extent to another causes higher disk rotations, and this is called Extent Fragmentation. SELECT OBJECT_NAME(OBJECT_ID), index_id,index_type_desc,index_level, avg_fragmentation_in_percent,avg_page_space_used_in_percent,page_count FROM sys.dm_db_index_physical_stats (DB_ID(N'AdventureWorksLT'), NULL, NULL, NULL , 'SAMPLED') ORDER BY avg_fragmentation_in_percent DESC
- 8. avg_fragmentation_in_percent: This is a percentage value that represents external fragmentation. For a clustered table and leaf level of index pages, this is Logical fragmentation, while for heap, this is Extent fragmentation. The lower this value, the better it is. If this value is higher than 10%, some corrective action should be taken. avg_page_space_used_in_percent: This is an average percentage use of pages that represents to internal fragmentation. Higher the value, the better it is. If this value is lower than 75%, some corrective action should be taken. Reducing fragmentation: Reducing Fragmentation in a Heap: To reduce the fragmentation of a heap, create a clustered index on the table. Creating the clustered index, rearrange the records in an order, and then place the pages contiguously on disk. Reducing Fragmentation in an Index: There are three choices for reducing fragmentation, and we can choose one according to the percentage of fragmentation: If avg_fragmentation_in_percent > 5% and < 30%, then use ALTER INDEX REORGANIZE: This statement is replacement for DBCC INDEXDEFRAG to reorder the leaf level pages of the index in a logical order. As this is an online operation, the index is available while the statement is running. If avg_fragmentation_in_percent > 30%, then use ALTER INDEX REBUILD: This is replacement for DBCC DBREINDEX to rebuild the index online or offline. In such case, we can also use the drop and re-create index method.