Run Rebuild Index Task daily on database but about 77 tables still highly fragmented over 80% !!!

Question

Hello everyone
On our particular database server, we run the Rebuild Index Task (Using classic Maintenance Plan Designer) every night. Running the  script below, I saw that about 77 tables had an avg_fragmentation_in_percentage between 80% and 99% !!
SELECT OBJECT_NAME(ind.OBJECT_ID) AS TableName,
ind.name AS IndexName, indexstats.index_type_desc AS IndexType,
indexstats.avg_fragmentation_in_percent
FROM sys.dm_db_index_physical_stats(DB_ID(), NULL, NULL, NULL, NULL) indexstats
INNER JOIN sys.indexes ind
ON ind.object_id = indexstats.object_id
AND ind.index_id = indexstats.index_id
WHERE indexstats.avg_fragmentation_in_percent > 30--You can specify the percent as you want
ORDER BY indexstats.avg_fragmentation_in_percent DESC
I dont understand why these tables are highly fragmented after a daily index rebuild! Unless the users are doing heavy inserts/updates/deletes during the day.
Anyone has an idea of the possible causes of these results??
Thank you all in advance

Answer 1

Hallo Efyuzegeekgurl,
this is a normal behaviour when your table is only a small one. The reason for that behaviour is the following:
The dmv sys.dm_db_index_physical_stats counts each page which is not located at the very next one of the same table as a "fragment". If the first data page is 1700 and the second one is located on 1705 SQL server will count it as a fragment.
The basic allocation algorithm of Microsoft SQL Server is that new data pages will be created in "mixed extents". An extent is a contigious line of 8 consecutive data pages. This is for historical reasons when storage was expensive and data should
use as much as possible of allocated storage.
If a new table will be created the very first 8 pages are not located one after the other but "could" be located on different extents. If your table has only 9 data pages than 8 pages will count as "fragmented" because of their non consecutive
allocation. After 8 pages have been allocated Microsoft SQL Server will ALWAYS use a full extent for the next 8 pages (and so on!).
To understand this behaviour use the following example which creates a table with a record length of 8K. The next command will create 10 records (which means 10 data pages).
CREATE TABLE dbo.foo
Id INT NOT NULL IDENTITY (1, 1),
c1 CHAR(8000) NOT NULL DEFAULT ('filler')
GO
SET NOCOUNT ON;
GO
INSERT INTO dbo.foo DEFAULT VALUES
GO 10
When you check the page allocation you'll find the first 8 data pages not in a consecutive row but "scrambled" into different pages:
SELECT database_id,
index_id,
partition_id,
allocation_unit_type_desc,
extent_page_id,
allocated_page_page_id,
is_mixed_page_allocation,
page_type_desc
FROM sys.dm_db_database_page_allocations
DB_ID(),
OBJECT_ID('dbo.foo', 'U'),
0,
NULL,
'DETAILED'
) AS DDDPA
WHERE
DDDPA.is_allocated = 1
ORDER BY
DDDPA.is_iam_page DESC,
DDDPA.page_level DESC;
GO
You will see that the first 8 pages are located in a MIXED extent. Although the pages are in a consecutive order in the extent 221112 it is not guaranteed!
You can see from the very first three records that they are not in a consecutive row stored.
Now the fragmentation of the data seems to be quite high because we have 4 breaks in the rows. If we compare this value to the "fragmentation in sys.dm_db_index_physical_stats we will see at least the same result:
SELECT index_type_desc,
alloc_unit_type_desc,
DDIPS.avg_fragmentation_in_percent,
DDIPS.fragment_count,
DDIPS.page_count,
avg_page_space_used_in_percent
FROM sys.dm_db_index_physical_stats
DB_ID(),
OBJECT_ID('dbo.foo', 'U'),
NULL,
NULL,
'DETAILED'
) AS DDIPS;
So - just from my point of view - don't count only to the fragmentation in percent because if the table is to small it will ALWAYS be over 30%. Take more into account the density of the pages and - as all others have mentioned - the number of pages of the
index.
I would consider indexes with more or equal to 1.000 pages.
Another tip: Forget about the f... maintenance plans. These "plans" are not worth the time and money because you cannot control it in a more precise way. Have a look to the solution of Ola Hallengren and implement his solution - a great master
piece of index maintenance:
http://ola.hallengren.com
All the best to the community :)
MCM - SQL Server 2008
MCSE - SQL Server 2012
db Berater GmbH
SQL Server Blog (german only)