Fragmentation 101
-
Comments posted to this topic are about the content posted at http://www.sqlservercentral.com/columnists/chedgate/fragm
-
One of the best things about books online is the free code.
If you look up dbcc showcontig you will see a routine on the help page that retrieves the table output of the dbcc and then runs index defrag against any index fragged at or above 30%
this is easily modified to be a rebuild or a reindex.
-
I have a server where there is heavy file fragmentation but it is quite busy and cannot be down for long periods of time. Is there any way other than shutting down SQL Server that would allow me to run a defrag on the datafiles while the the database files are in use?
-
Not any way that I can think of. Either shut down SQL Server or detach that specific database while you are defragging the files, otherwise it will not work.
-
One thing that you forgot to mention, when rebuilding the indexes you can specify a fill factor for the index. This is helpful if you have a clustered index where the new record will be inserted between existing records. Not as helpful, when all new records are inserted at the end of the table.
-
Hello!!!
I have been in trouble with my SQL Query that was causing time out on the UI...Investigation with DBCC showcontig reveiled that the Scandensity for these tables are 50% or less.
But to my surprise the Scandensity was not improving with REINDEXing and DEFRAG....(Number of rows in these tables are a few hundred only) . I am not sure why is this happening....
to some how escape I tried moving these tables to new ones and saw some improvement in the Scandensity (Query to a improved slightly).
But what puzzles me is why REINDEXing and DEFRAG did't work????
Also one more question....
If I make a backup of this database and restore into a new server will that create new data/index pages?? does that have the same fragmentation as when it was backed up??? or does the RESTORE just uncompress the old pages???
Thanks
Cheriyan.
-
DisKeeper claims to be able to defrag locked files.
I've used it a couple times in production.
A very nice feature it has is that it monitors the IO queue. If the queue rises, it backs off and allows the system more time on the disks. The effect is zero (Near zero) impact while doing a defrag. The con is your defrag could take forever. But you have the time...
Cheers,CrispinI can't die, there are too many people who still have to meet me!It's not a bug, SQL just misunderstood me!
-
On a whole, a good article.
One correction though.
When CREATE INDEX, WITH DROP_EXISTING is executed on a clustsered index, all non clustered indexes have to be rebuild.
A NC index contains keys to the clustered index, i.e. the data. When you rebuuild this, the data could, and probably will, move to another page. This would cause the key link to become invalid.
See BOL:
CREATE INDEX, DROP_EXISTING
"Because nonclustered indexes contain the clustering keys, the nonclustered indexes must be rebuilt when a clustered index is dropped. If a clustered index is recreated, the nonclustered indexes must be rebuilt to take the new set of keys into account"
I learnt this the hard way...
Cheers,CrispinI can't die, there are too many people who still have to meet me!It's not a bug, SQL just misunderstood me!
-
To try and answer your first question:
If you only have a few rows (couple could be a few) in a table, SQL creates those pages on an extent which is shared by other pages. An extent is 64k, page 8. If you have a table with narrow rows and only a couple hundred rows, the chance is good you using a mixed extent. With this in mind, SQL may not defrag it because a) it's so small, b) there's other data sitting on the extent.
Try adding a couple thousand rows to the table (Assuming you can) and see if that changes things.
See BOL, pages and extents for more info.
To your second question, yes.
When SQL does a restore, it restores all tables. One of them being sysindexes. Sysindexes determines on what pages / extents sit. DUring a restore, SQL acquires a chuck of disk space. It then creates all the extents / pages. Once done, it starts writing out the data page by page. (or extent by extent?).
Restoring a backup will only give you contiguous OS space, not within the SQL files.
Cheers,CrispinI can't die, there are too many people who still have to meet me!It's not a bug, SQL just misunderstood me!
-
Hm, now I do not have the experience from learning the hard way about this (not that I can recall at least). However, you did not copy-paste all what BOL says:
"The DROP_EXISTING clause enhances performance when re-creating a clustered index (with either the same or a different set of keys) on a table that also has nonclustered indexes. The DROP_EXISTING clause replaces the execution of a DROP INDEX statement on the old clustered index followed by the execution of a CREATE INDEX statement for the new clustered index. The nonclustered indexes are rebuilt once, and only if the keys are different.
If the keys do not change (the same index name and columns as the original index are provided), the DROP_EXISTING clause does not sort the data again. This can be useful if the index must be compacted."
To me, this seems to say that if you do not change the keys (e.g. run this simply to defrag the index) then the NC indexes do not need to be rebuilt at all. Even if the keys do change, the NCs only need to be rebuilt once instead of twice (one for DROP INDEX and one for CREATE INDEX). This last part is something that my article should have mentioned.
Also, this is not correct:
A NC index contains keys to the clustered index, i.e. the data. When you rebuild this, the data could, and probably will, move to another page. This would cause the key link to become invalid.
If the keys of the clustered index stay the same then the 'links' in the NCs are not invalid even if rows move to different pages. That is the whole idea of the clustered index key, instead of the NCs pointing directly to a physical oage (with a RID), they store the key that can be used to seek the clustered index to find the row.
-
Good article! Really liked the fact that it clearly indicated problems and solutions (to include pros and cons). Thank you!
-
We recently migrated from SQL 7 to SQL 2000. In SQL 7 whenever a clustered index was rebuilt the NC would get rebuilt, but in SQL 2000 i do not see the NC being rebuilt. Atleast not in last 2 weeks of what i have seen. So is this an enhancement in SQL 2000 or does that mean the keys havn't changed to affect rebuild of NC?
Very good article and the dicussion is also providing lot of information.
-
The defrag that comes with win2k3 is capable of defragging the files while sql server is online, it's a huge I/O hit and the server is pratically unusable; it's possible though.
-
We have our SQL databases on RAID sets on a SAN. I assume, therefore, that the disk defragmentation discussion is moot, since files on this hardware are fragmented by design. I wouldn't know where to begin to figure out which disks and how contiguous the files were.
Given that, one shouldn't spend too much time trying to optimize table or index storage either, other than having fillfactors that are consistent with transaction load, right?
Thanks
-
I have read an article on physical file defragmentation on a SAN or RAID system http://www.raxco.com/products/perfectdisk2k/whitepapers/pd_raid.pdf.
The gist of RAID/SAN defrag is that
a) Not all defrag tools are compatible with SANS/RAID arrays.
b) It is the LCN numbers that get defragged rather than the physical files and this improves performance.
There is a benefit in performance of having the data striped across the disks in a RAID/SAN but the LCN benefit from having contiguous blocks.
Has anyone got any practical experience with this?
Viewing 15 posts - 1 through 15 (of 24 total)
You must be logged in to reply to this topic. Login to reply