SQLUninterrupted

I am just a medium, SQL Server the Goal

Optimizing your backup – Tips and Tricks

Posted by Sourabh Agarwal on September 25, 2013

Last week at the TechCon2013, sponsored by the SQLBangalore User Group and other User Groups in Bangalore, I did a talk on how to optimize your backup/restore. I focused on the three main aspects of Backup/Restore and talked about some of the things which can be done to optimize them

  1. Read from Disk
  2. Store in Memory
  3. Write to Disk/Tape

For the Demos, I used 3 databases as described.

Database 1

  • Size – 12 GB
  • DB Name – AdventureWorks2012SingleFile
  • Number of Data Files: 1
  • Disk Allocation Size: 4KB

Database 2

  • Size – 12 GB
  • DB Name – AdventureWorks2012MultiFile
  • Number of Data Files: 4 (spread on 4 different physical disk)
  • Disk Allocation Size: 64KB

Database 3

  • Size – 4 GB
  • DB Name – AnotherSingleFileDatabase
  • Number of Data Files: 1
  • Disk Allocation Size: 64KB

The slide deck for the presentation is included below.

Optimizing Reads

To optimize reads, I focused mainly on the following

  1. Reading from a single MDF file vs. reading from multiple Data files
  2. Using higher disk allocation unit, I would be writing another post on effect of disk allocation unit size to SQL backup.

All backups were performed to a Null device. The idea was to test the read performance. I saw the following backup performance for the databases mentioned above.

   1:  

   2: backup database [AdventureWorks2012SingleFile]

   3: to disk = 'NUL' WITH COPY_ONLY

   4: --BACKUP DATABASE successfully processed 1521753 pages in 26.510 seconds (448.460 MB/sec).

   5:  

   6: backup database [AnotherSingleFileDatabase]

   7: to disk = 'NUL' WITH COPY_ONLY

   8: --BACKUP DATABASE successfully processed 569482 pages in 9.202 seconds (483.489 MB/sec).

   9:  

  10: backup database [AdventureWorks2012MultiFile]

  11: to disk = 'NUL' WITH COPY_ONLY

  12: --BACKUP DATABASE successfully processed 1599457 pages in 25.244 seconds (494.998 MB/sec).

I get better performance with later disk allocation sizes and multiple sizes.

Optimizing Writes

To optimize reads, I focused mainly on the following

  1. Writing to a single backup file vs. writing to multiple backup files
  2. Using Compression
  3. Changing the MaxTransferSize option for the backups.

All backups were performed to a 4K allocation unit drive, except for the multiple file backup, which were performed on a 64K Allocation unit disk. The following backup performance was observed.

   1: backup database [AnotherSingleFileDatabase]

   2: to disk = 'C:\AnotherSingleFileDatabase.bak'

   3: Go

   4: --BACKUP DATABASE successfully processed 569493 pages in 52.510 seconds (84.729 MB/sec).

   5:  

   6: -- Delete the file to preserve Space

   7: xp_cmdshell 'del C:\AnotherSingleFileDatabase.bak'

   8: -- Backup taken to Multiple Files on Different drives of 64K disk allocation Units

   9: backup database [AnotherSingleFileDatabase]

  10: to 

  11: disk = 'F:\AnotherSingleFileDatabase_BackupFile1.bak',

  12: disk = 'G:\AnotherSingleFileDatabase_BackupFile1.bak',

  13: disk = 'H:\AnotherSingleFileDatabase_BackupFile1.bak',

  14: disk = 'E:\AnotherSingleFileDatabase_BackupFile1.bak'

  15: Go

  16: --BACKUP DATABASE successfully processed 569490 pages in 33.600 seconds (132.414 MB/sec).

  17: --- Delete all the files

  18: xp_cmdshell 'del E:\AnotherSingleFileDatabase_BackupFile1.bak'

  19: go

  20: xp_cmdshell 'del H:\AnotherSingleFileDatabase_BackupFile1.bak'

  21: go

  22: xp_cmdshell 'del G:\AnotherSingleFileDatabase_BackupFile1.bak'

  23: go

  24: xp_cmdshell 'del F:\AnotherSingleFileDatabase_BackupFile1.bak'

  25: go

  26:  

  27: -- Backup With COMPRESSION

  28: backup database [AnotherSingleFileDatabase]

  29: to disk = 'C:\AnotherSingleFileDatabase.bak'

  30: WITH COMPRESSION

  31: Go

  32: ---BACKUP DATABASE successfully processed 569491 pages in 25.721 seconds (172.977 MB/sec).

  33: xp_cmdshell 'del C:\AnotherSingleFileDatabase.bak'

  34: Go

  35:  

  36: -- MAXTRANSFER OPTION

  37: DBCC TRACEON(3605,3213,-1)

  38: -- Writes backup Configuration related informtion to the trace file.

  39:  

  40: backup database [AnotherSingleFileDatabase]

  41: to disk = 'C:\AnotherSingleFileDatabase.bak'

  42: WITH COMPRESSION

  43: go

  44: xp_cmdshell 'del C:\AnotherSingleFileDatabase.bak'

  45: Go

  46: sp_readerrorlog

  47: go

  48:  

  49: backup database [AnotherSingleFileDatabase]

  50: to disk = 'C:\AnotherSingleFileDatabase.bak'

  51: WITH COMPRESSION, MAXTRANSFERSIZE = 2097152

  52: Go

  53: -- BACKUP DATABASE successfully processed 569490 pages in 21.920 seconds (202.971 MB/sec).

  54: xp_cmdshell 'del C:\AnotherSingleFileDatabase.bak'

  55: Go

We got better write performance when using a 64K allocation unit disk and when using Multiple Files. Further performance improvement was observed with Compression and by increasing the MaxTransferSize for the backups.

Optimizing Memory

To optimize reads, I focused mainly on changing the BufferCount option for the backups. This would help create more buffers in the memory for the buffer.

Total Memory used = BufferCount * MaxTransferSize

While i did not have a demo for the performance improvement when increasing the number of buffer, I had a demo on the side affect of increasing the BufferCount/MaxTransferSize to be very high.

   1: backup database [AnotherSingleFileDatabase]

   2: to disk = 'C:\AnotherSingleFileDatabase.bak'

   3: WITH COMPRESSION, MAXTRANSFERSIZE = 2097152, BUFFERCOUNT = 3000

   4: Go

   5: -- Error Message

   6: --There is insufficient system memory in resource pool 'default' to run this query.

If we increase the BufferCount and the MaxTransferSize to be very high, we would get into memory issues on the server.

Hopefully this information helps.

Posted in Backup, Performance, SQL Engine, SQL Server, Storage Engine | Leave a Comment »

TECHCON September 21 2013 – A Day to Remember

Posted by Sourabh Agarwal on September 24, 2013

September 21st 2013, marked a remarkable day for the SQLBangalore UG, BITPro, PSBUG (Power Shell Bangalore User Group) and BDotNet UG, when all the four groups came together to host one of the most extensive technical conference ever in Bangalore.

With over 250+ Participants and 20 session being conducted in parallel across three conference rooms at the Microsoft facility at Bangalore, this easily was the biggest technical conference brought to you by the community. I had the privilege to be part of the SQL Server Track, both as the host and the presenter and it was a pleasure to see close to 130 participants coming in as early as 8:30 AM in the morning on a Saturday, which speaks volumes about their passion and the desire to learn.

The Day started on a very healthy note with Ryan Fernando (Founder:Qua Nutrition L|B) speaking on “Techies Guide to Better Nutrition”. Ryan’s talk was really an Eye opener and with my wife accompanying me, it opened up the Pandora’s box. I don’t think I need to mention what kind of treat, I am in for later when we reach home.

The second session was on Big Data, the hottest commodity on the Database market. Amarpreet Basan, Technical Lead, Microsoft SQL Server Support team at IGTSC talked in details about Big Data and did some really cool demo’s with HDInsight and Hadoop clusters. The slide decks and the demo’s scripts for this would be available on the SQL Bangalore UG page on Facebook.

I had the privilege to present next (this being my 4th session for the User Group). For this session I went with something which we always talk about in Best Practices, but not necessarily follow it in when working with SQL Server. I focused mainly on the Optimizing your Backups, where in we talked about how to optimize the backup operation for your SQL Server.

Prabhjot Kaur, (IGTSC, SQL Support) went next with her session on demystifying the myths of Tempdb. All the participants present work with SQL Server day in and day out, and would have spent a considerable time optimizing Tempdb for their SQL Server environments and the important of this presentation could be estimated from the fact that here was pin drop silence in the room for the 50 odd minutes when Probhjot was talking.

Next in line was the most important part of the day, LUNCH. A highly nutritious lunch was provided by iTIFFIN (No Its Not An APPLE product). With 499 Calories and absolutely the right ingredients, this was exactly what the doctor ordered for the DBA/Developer crowd in the room.

To make lunch even more interesting we had Balmukund Lakhani (B|T|F) with his usual trivia’s and jokes. All in all the 45 minute lunch break was both healthy and fun filled, thanks to ITiffin and coordinators for this event.

Post lunch we had 3 sessions lined up on some of the most sought after topics in the Market right now. We had Ajay Kumar (Technical Lead, IGTSC) and Sunil Kumar B.S (Escalation Engineer IGTSC) talking about SQL Azure, the Microsoft Cloud offering for SQL. Their session slide deck and the demo content would be made available on the SQL Bangalore User Group page on facebook.

This was followed by Sumit Sarabhai’s (Support Escalation Engineer, IGTSC) session on Query Tuning and Optimizations. With a room full of SQL Developers and Administrators, there was no doubt that this would end up as the most saught after sessions. I mean, is there a SQL Developer/Admin who hasn’t had the bad luck of having to tune a query and make it run faster. Sumit’s session covered some of the salient points about what are the things to do when tuning a query.

To end the sessions, we had Selva R. (Sr. Support Escalation Engineer, IGTSC), presenting on one of the coolest (if not the coolest) thing in the MSBI stack, Power BI. Speaking about Power BI, always reminds of a line in the Second Spiderman Movie (starring Tobey Maguire): “The Power of sun, in the palm of my hands”. Such is the power of Power BI, the master set comprising of Power View, Power Pivot, Power Map and Power Query.

We finished the day around 5:30 PM, with Balmukund fittingly pulling the curtains on a day filled with learning and knowledge sharing. After having worked the entire week and then having spent another 9 hours on a weekend, I thought I have had it enough.I could not have been more wrong, it was heartening to see most of the folks hanging around to interact and sharing their SQL Server experience with one another.

To sum it up, it was an over whelming experience being part of such a fabulous event, and the only parting words I could think off is “May the SQL Server be with you

Posted in SQL Engine | 4 Comments »

SQL Server Record Structures–Part 4

Posted by Sourabh Agarwal on October 25, 2012

Previous Posts in this series.

  1. SQL Server Record Structures-Part 1
  2. SQL Server Record Structures-Part 2
  3. SQL Server Record Structures-Part 3

In the pervious posts I have discussed the structure of the data record in SQL (without compression or sparse columns). We also discussed a few special cases which affect how the record is stored in SQL. In the ensuing post, I would be talking about record structures when data compression is enabled on the table/index.

Data compression was first introduced in SQL with SQL Server 2005, in the form of VarDecimal Storage, where in any column with Decimal/Numeric data type was converted into VarDecimal storage in order to utilize only the required amount of bytes. This obviously was subject to enabling VarDecimal Storage on the DB and the table.

With SQL Server 2008, two new compression techniques were introduced. VarDecimal Storage was deprecated in this version of SQL as row compression achieves the almost the same results.

Row Compression

When Row compression is enabled on the table/Index all columns in the table are converted to use variable length storage, using only enough bytes to store the data. For example if a column is declared as Char(50) and has the following string as data “SQLUninterrupted” instead of using the 50 bytes of storage( as would be the case when no compression is enabled), SQL only uses 16 bytes when compression is enabled.

Some points to keep in mind when enabling Row Compression on a table/Index

  1. Compression does not change the Max Size limits on the record or Index keys.
  2. Compression cannot be implemented if the maximum size of the record, plus the Compression Information overhead exceed 8060 bytes.
  3. Compression can be enabled on individual partitions of a Partitioned table/Index.
  4. Changing the compression setting of a heap table requires a rebuilt of all the non-clustered indexes on the table.

Page Compression

When Page Compression is enabled on a Table/Index it takes a 3 level approach. First row compression is enabled on the page, followed by a Prefix and Dictionary compression. Details of how Prefix and dictionary compression work is available in Book Online topic “Page Compression Implementation”. When Page Compression is enabled, a special record called the Compression Information Record is added to the page. This CI record stores the Prefix/Dictionary compression information.

Some points to keep in mind when enabling Page Compression on a table/index.

  1. Page Compression only kicks in when the Page is full and a new row is being inserted. If enough space can be saved to accommodate the new record and the compression information on the page, the page will be compressed. Otherwise it would not be compressed.
  2. Enabling Page Compression on a heap table does not enforce compression on the pages until the heap is rebuilt.
  3. New pages being added to the heap are also not compressed until the heap is rebuilt.
  4. Non-Leaf pages of an index cannot be page compressed. This is to avoid the overhead of uncompressing the record during index operations.

Compression Record Structure

When compression is enabled on a table, the data records are stored in a new format commonly called the CD (Column Descriptor) format. This format is entirely different from the regular record structure used when there is no compression or no sparse columns defined in the table. The Structure of the record takes the following format.

image

Record Header Region

This is a 1 byte section with the following information

  • Bit 0 – Type of Record. This is set to 1 for CD format Records
  • Bit 1 – Indicates if the record has versioning information.
  • Bit 2-4 – Indicates the formation stored in the row. Details of these bits can be found in the Microsoft® SQL Server® 2008 Internals by Kalen Delaney, Paul S. Randal, Kimberly L. Tripp , Conor Cunningham , Adam Machanic
  • Bit 6 – Indicates if the record contains a long data region or not.
  • Bit 7 – Not used.

Column Descriptor Region

This region is composed of two parts, the first part is either 1 byte or 2 bytes and indicated the number of columns in the table. The second part is 4 bit (per column) descriptor which indicates the length of the column. The 4 bits can be used to represented 16 different values, of which only 13 are used with SQL 2008. For details of the descriptors refer the SQL Server 2008 Internals book mentioned above.

image

Short Data Region

Contains columns which are less than equal to 8 bytes in length. Since the length of the columns are not directly stored in this region, in tables with a lot of columns, it can be expensive to access the last column, as this would require processing all the previous columns (length of each column) and then computing the start and end of the desired column. To mitigate this, the columns are divided into clusters of 30 columns each. A Short Data Cluster Array(byte array) is then created in this region to store the length of each cluster. Since there are only 30 columns in each cluster (with each column being max of 8 bytes), a single cluster can have a max length 240 bytes.

image

Long Data Region

All columns with data length grater than 8 bytes are stored in the long data region. Since the length of these columns is not stored in the CD regions (each CD descriptor is only 4 bits, whereas the column length can be anything), the Long Data region requires a column offset array to correctly locate the column data in the Long Data Region.

Again, in order to reduce the cost of locating a column value in this region, we have a Long Data Cluster array, which  stores the number of columns in each clusters (again each cluster represents up to 30 columns).

image

Note: This illustration has been taken from the Microsoft® SQL Server® 2008 Internals by Kalen Delaney, Paul S. Randal, Kimberly L. Tripp , Conor Cunningham , Adam Machanic

Special Information Section

Contains optional information about forwarding/back pointer or the Versioning information.

Row Compression Example

Create Table RowCompressionExample

(

Col1 int, 

Col2 bigint, 

Col3 char(40), 

Col4 char(30), 

col5 numeric(18,7), 

Col6 varchar(300), 

col7 datetime, 

col8 varchar(400), 

col9 char(100), 

col10 char(100)

)

 

-- Enable Row Compression on the table

ALTER TABLE RowCompressionExample REBUILD WITH (DATA_COMPRESSION = ROW)

 

----- Insert 2 records into the table using

insert into RowCompressionExample values

(10, 345678345, 'Sourabh', 'Agarwal', 123345456.3456, 'This is first Long data', getdate(), 

'This is a Second long Data', 'This is a third long Data', 'short')

 

insert into RowCompressionExample values

(76854, 2000, 'Short1', 'LongDataValue1', 3847.34, 'This is first Long data', getdate(), 

'This is a Second long Data', 'This is a third long Data', 'LongDataRegion')

 

-- Use DBCC IND to display the Pages used for this Table.. 

DBCC IND(10,'RowCompressionExample',-1)

 

--- Use DBCC PAGE to check the records structure on the Page

DBCC TRACEON (3604, 1)

DBCC PAGE(10,1,285,3)

I have created a sample table with 10 columns of varying size and then enabled Row Compression on the Table. I then inserted two records in the table. Now Lets look at the record structure from the output of the DBCC Page command.

image

As indicated by the DBCC PAGE output, the total record length is only 128 bytes as compared to 362 bytes that might have been required to store the same record in the uncompressed format.

Next we also have the CD Array displayed in the output. Since there are 10 Columns in the table, there are 10 entries in the CD array.

As can be seen, the columns Col6, Col8 and Col 9 have values longer than 8 bytes and hence are marked as Long in the CD array. Also notice that since there are only 10 columns, all the columns are part of the same cluster (cluster 0).

Page Compression Example

I used a similar table structure  as above, to demonstrate Page Compression. As I mentioned earlier, Page compression does not kick in until the Page is full and a new record has to be inserted into the table.

I created the following tabled and enabled Page Compression on the Table.

Create Table PageCompressionExample

(

Col1 int, 

Col2 bigint, 

Col3 char(40), 

Col4 char(30), 

col5 numeric(18,7), 

Col6 varchar(300), 

col7 datetime, 

col8 varchar(400), 

col9 char(100), 

col10 char(100)

)

 

-- Enable Row Compression on the table

ALTER TABLE PageCompressionExample REBUILD WITH (DATA_COMPRESSION = PAGE)

Next I inserted 64 records into the table using the following script

declare @count int = 0

While(@count < 32)

begin

insert into PageCompressionExample values

(10, 345678345, 'Sourabh', 'Agarwal', 123345456.3456, 'This is first Long data', getdate(), 

'This is a Second long Data', 'This is a third long Data', 'short')

insert into PageCompressionExample values

(76854, 2000, 'Short1', 'LongDataValue1', 3847.34, 'This is first Long data', getdate(), 

'This is a Second long Data', 'This is a third long Data', 'LongDataRegion')

set @count = @Count+1

end

At this point, there are two data pages in the table with row compression only. So why did the Page Compression not take effect?

The reason is that on a heap, PAGE Compression would not take effect until the heap table has been rebuilt. So lets rebuild the table..

ALTER TABLE PageCompressionExample REBUILD WITH (DATA_COMPRESSION = PAGE)

Now there is only one Page in the table with Page Compression enabled..

When Page Compressed, SQL Server adds a new record to the top of the page (right after the Header Info) called the Compression Information record. Additionally SQL server adds information in the page header to indicate that the page is “Page Compressed”. If the value of m_typeFlagBits = 0x80, then the page is Page Compressed.

In the next part of this blog series I will talk about the Compression Information record structure.

Posted in Record Structures, SQL Engine, SQL Server, Storage Engine | Tagged: , , | Leave a Comment »

SQL Server Record Structures–Part 3

Posted by Sourabh Agarwal on October 19, 2012

In the previous posts in this series I had discussed about Regular Data Record structures and the special case of Row Forwarding. In this post I will talk about two other special cases and how the record structures are changed or impacted with they are present.

  • Row Versioning
  • Ghost Records/Ghost Cleanup

Row Versioning

Row Versioning was first introduced in SQL Server 2005. Several SQL Server features (Online Index Rebuild, CheckDB, etc.) and the 2 new row versioning based Isolations Levels (Snapshot Isolation and Read Committed Isolation level) in SQL use Row Versioning. There is a plethora of content on MSDN and other blogs about how the row versioning based Isolations work, it is for this very reason I wont be talking about these features in this post. Instead, what I would concentrate is how row versioning based isolation levels change the record structure in SQL Server.

Row Versioning is applicable to all types of records (data, Index, text) in SQL. When any of the row versioning based Isolation level is enabled on the database, for any update on an existing record in the table, SQL Server creates a last committed copy of the record and put this copy in the version store in the Tempdb database. Any other operation that attempt to read this record (i.e. before the update transaction has committed) would read the record from the version store.

The original record, which remains in the database is modified to append a 14 Byte Versioning Tag. This 14 byte is used to present the Time Stamp of when the record was created and a pointer to the pervious version of the record in the Version Store.

Example

--- create a table for testing for the versioning, insert two records in the table. 

 

Create Table VersionedRecords (Col1 int Identity(1,1), Col2 varchar(100), Col3 Datetime, Col4 char(50))

go

insert into VersionedRecords(col2, col3, Col4) values (Replicate ('Sourabh', 10), getdate(), Replicate('AAA', 10))

go

insert into VersionedRecords(col2, col3, Col4) values (Replicate ('Agarwal', 14), getdate(), Replicate('BBB', 10))

go

 

-- Now lets check the Record sizes for this Table.

 

DBCC IND ('DatabaseName', 'VersionedRecords',-1)

DBCC PAGE(10, 1, 281, 3)

the Record Sizes are as below….

image

The sizes for the records 143 and 171 bytes respectively, which is in accordance with what I discussed in the part 1 of this blog series.

Now lets enable the Snapshot Isolation on the database.

-- Enable Snapshot Isolation in the DataBase.

Alter Database DatabaseName Set ALLOW_SNAPSHOT_ISOLATION  ON

Once the Database option was enabled, I ran an Update Statement on the Table using another connection.

-- Run the Update transaction so that the Versioned Records are created

begin transaction 

update VersionedRecords set Col2 = Replicate('Sourabh', 14) where Col1 = 1

now if we dump the data page again, we see that the record size for the 1st record changes, while the one for 2nd record remains the same.

image

we can see that there is a increase of 42 bytes in the record (28 extra bytes to store the new value + 14 bytes for the versioning tag). details of the versioning information (14 byte) will be discussed in another blog at a later time.

Side Pointers

  1. Existing records in a table are not immediately modified when Row Versioning is enabled on the data, they are only done when a subsequent update operation happens on the record.
  2. All new records added to the table (after changing the database setting) would have the Versioning bytes appended to them.
  3. Versioning store is located in the tempdb, and can cause of I/O activity in the TempDB. This can lead to performance issues on the server, if Tempdb storage is not carefully planned.
  4. Multiple updates can be done on the same record which can lead to a chain of versioning records. Any subsequent read operation has to traverse this chain to get the correct version of the record. This can be time consuming and hence can lead to performance issues.
  5. Version store cleanup is a background process and there can be scenarios where the rate of cleanup is less than the rate of generation of version records. This can lead to increased Tempdb usage.

Ghost Records/Ghost Cleanup

In SQL Server when a record is deleted from the table, the physical storage for the record is not immediately destroyed or deleted, instead the record is marked for deletion and the actual deletion happens at a much later point in time. These records which are marked for Deletion are called Ghost Records. This process significantly increases the performance of the delete operations. How? I will leave it up to you to figure that out.

In the 2 byte record header, the combination of bits 1-3 is used to represent whether the record is ghost record or not.

Bits 1 through 3 Taken as a three-bit value

5 indicates a ghost index record,

6 indicates a ghost data record, and

7 indicates a ghost version record.

The actual deletion of the records is performed by the Ghost Cleanup Task, which runs at an interval 10 seconds (SQL 2008 and above) and 5 seconds (SQL 2005 and below). The deleted records are added to the delete queue of the ghost cleanup task, during a table scan.

Another important thing to remember is that the records are not really deleted, just the space is marked as not being used.

Posted in Record Structures, SQL Engine, SQL Server, Storage Engine | Tagged: , | Leave a Comment »

SQL Server Record Structures–Part 2

Posted by Sourabh Agarwal on October 17, 2012

In the first part of this series, I had discussed about the regular record structure used in SQL Server. In this post I would be talking about a special case, Row Forwarding and how they effect the record structures in SQL.

In order to understand Row Forwarding, it’s important to first understand how non clustered indexes created on a Heap table works.

in case of heap tables, non-clustered indexes on their leaf pages have the RID along with the index key values. This RID value helps link the non-clustered index to the heap table during a scan or a seek. Consider the following example. This RID value is combination of PageID:RowID identifying a physical record in the table.

Assume we have a heap table with the following columns (Col1, Col2, Col3, Col4, Col5). Also assume there is a non-clustered index on the column (col1).

The index leaf page would have a similar structure (this is just an illustration!)

Index Key(a)  RID
1 222:0
2 222:1
3 222:2
4 222:3
5 222:4
6 222:5
7 222:6

 

now assume if we are running the following query, against the table

Select Col1, Col2, Col3 from Table1 

where Col1 = SomeValue

this query would can potentially use the Non Clustered Index (depending on whether the cost of using the index is lesser than table scan on not). If the query uses the NCI, then it can get the values of Col1 quiet easily, but for the values of the columns Col2 and Col3, it has to piggy back on the RID value to reach the actual data record in the table (PageID and RowID) and get the values from there.

This operation in SQL is called the Bookmark Lookup (SQL 2000) or the Lookup Operation (SQL 2005 and onwards).

Now assume we have a table with some Variable length columns. When the values in the variable length columns are updated, the update might result in an increase in the size of the column. SQL server might not be able to fit this new record on the same page and may cause a Page Split, thereby moving the current record and potentially other records on the page to a different page.

Now if there was a Non-clustered index on this table, then the non-clustered index would have to able to be modified to reflect movement of the rows. This would make the update operations very expensive.

So instead of having to update the non-clustered index, SQL server creates a pointer/stub at the initial location of the record to point to the new address of the record. This way, when the NCI scan or seek reaches the record, it simply reads the pointer record and reach the new location of the Row. The pointers are called forwarding Pointers/records and the actual record is called the Forwarded Records.

The same record can be modifies multiple times and a new forwarded record might have to be created. This could potentially lead to having a chain of forwarding/forwarded records. In reality this does not happen. What actually happens is that the Forwarded Record also contains a back-pointer to the forwarding record. So when the multiple changes are being made to the record, the engine just takes the new location of the record and updates the original forwarding record, to point to this new location.

Row Forwarding is bad for performance. Also, row forwarding only happens in HEAP Tables.

Example: Examining the Forwarding Record and Forwarded Record

CREATE TABLE ForwardingRecord

(

Col1 int NOT NULL,

Col2 char(1000) NOT NULL,

Col3 varchar(3000) NULL,

Col5 varchar(4100) NOT NULL

);

 

Insert into ForwardingRecord values (1, Replicate('a',1000), replicate('b',1000),replicate('b',1000))

Insert into ForwardingRecord values (2, Replicate('a',1000), replicate('b',1000),replicate('b',1000))

 

DBCC IND('DatabaseName','ForwardingRecord',-1)

 

DBCC PAGE(10,1,228,3)

The result of the DBCC Page, would show that there are two records on the page. Each record is about 3017 bytes in size.

now lets update the second record in just a way that it causes a Page Split.

Update ForwardingRecord set Col5 = Replicate('v', 4100) where col1=2

Dumping the same page again and looking at the Slot 1 (the original record was here) we see,

image

As can be seen, the forwarding record has the information about where the forwarded record exists. Let’s now try to dump the Page 280 and check the record.

image

as can be seen that the Forwarded record has information about the forwarding record. As Paul Randal mentions in this blog, the back pointer is 10 bytes in size.

Posted in Record Structures, SQL Engine, SQL Server, Storage Engine | Tagged: , , | Leave a Comment »

 
Follow

Get every new post delivered to your Inbox.

Join 82 other followers