Skip to main content

Predicting the performance of database (table/index) compression

SQL Server row or page level compression is a very handy option in many cases. However it will be nice if we can predict what will be exact space saving before we do any change in a production table. In other way, this is an answer to the below question:

How can we estimate (predict) the savings from compression?

Use below SP to estimate the savings from SQL Server compression:

sp_estimate_data_compression_savings 
     [ @schema_name = ] 'schema_name'  
   , [ @object_name = ] 'object_name' 
   , [@index_id = ] index_id 
   , [@partition_number= ] partition_number 
   , [@data_compression= ] 'data_compression' 
[;]


Example:

We are testing the compression performance of table named “dbo.test” from database “testdatabase” and we are using page level compression on entire table.

use testdatabase
EXEC sp_estimate_data_compression_savings 'dbo','test',null,null,'page'

Example output:



object_name                                                                                                                      schema_name                                                                                                                      index_id    partition_number size_with_current_compression_setting(KB) size_with_requested_compression_setting(KB) sample_size_with_current_compression_setting(KB) sample_size_with_requested_compression_setting(KB)
-------------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------- ----------- ---------------- ----------------------------------------- ------------------------------------------- ------------------------------------------------ --------------------------------------------------
test                                                                                                         dbo                                                                                                                              1           1                57087872                                  28594376                                    40744                                            20408

From this example, the estimated compression is around 50% (28594376/57087872)

For more details on this stored procedure, please refer to http://msdn.microsoft.com/en-us/library/cc280574.aspx
Labels:
Location: Greer, SC 29650, USA

Comments

Post a Comment

Popular posts from this blog

How to kill a negative SPID (like SPID -2) in SQL Server?

Rarely this scenario will arise when most likely you see this negative SPID (most likely SPID -2) is blocking other transaction causing issues. If you try to kill it using normal KILL command, it will fail reporting below error: Msg 6101, Level 16, State 1, Line 1 Process ID <SPID Number> is not a valid process ID. Choose a number between 1 and 2048 This is because of an orphaned distributed transaction ID.  Please follow below steps to kill it: Step 1: -- Find the UOW Number select req_transactionUOW from master..syslockinfo where req_spid = <SPID Number> --  <SPID Number>  is -2 most likely. Step 2: -- Copy the UOW number from Step one KILL ‘<UOW Number>’ This will kill the negative SPID resolving the issue.  However please note following points: 1. For SPID -2, you may find multiple UOW numbers. Please start killing them one by one. Typically killing first UOW will resolve the issues. (ie. will kill all UOW and release
Post a Comment
Read more

DMV/TSQL to find out basic hardware information of the SQL Server including when SQL Server started.

Please use below code: However, please be advised that it can not tell correct information around virtualization.  For example, it will show Hypervisor even if SQL runs on a physical OS where Hyper-V is on. So use this query only when you do not have sufficient access on underlying Windows Operating system to get these information directly. -- Basic hardware information for SQL Server (sys.dm_os_sys_info) /* This query is courtesy of https://sqlserverperformance.wordpress.com/. All credits goes to original author. */ SELECT cpu_count AS [Logical CPU Count] , scheduler_count , hyperthread_ratio AS [Hyperthread Ratio] , cpu_count / hyperthread_ratio AS [Physical CPU Count] , physical_memory_kb / 1024 AS [Physical Memory (MB)] , committed_kb / 1024 AS [Committed Memory (MB)] , committed_target_kb / 1024 AS [Committed Target Memory (MB)] , max_workers_count AS [Max Workers Count] , affinity_type_desc AS [Affinity Type] , sqlserver_start_time AS [
Post a Comment
Read more

‘Trace Skipped Records’ – What is this and how to resolve it while using SQL Server Profiles?

In some very rare case, you may experience a very weired message in profiler’s output as ‘Trace Skipped Records’ while you trace something on SQL Server. Screenshot of similer situation is as below: This is not an error but it comes by design of SQL Server (I believe so). When you are using SQL profiler and return data is too big to fit in the GUI (for me, it is an enormous xml), SQL Server simply prints this message and proceed to next step. Mostlikely this is to save server’s memory and performance. Although not suggested and guranteed, you can try to run a server side trace and dump data in a file which should capture all the data. However, it is strongly not recommended to run a trace on your production server from server side. Microsoft will probally document this limitation in future. More details may be found at https://connect.microsoft.com/SQLServer/feedback/details/304225/msft-edw-profiler-displays-trace-skipped-records-for-large-530kb-batch 
Post a Comment
Read more