General SQL Server Performance Tuning Tips
When your transaction log grows large and you want a quick way to shrink it, try this option. Change the database recovery mode of the database you want to shrink from “full” to “simple,” then truncate the log file by performing a full backup of the database, then switch back to the “full” recovery mode. By temporally changing from the Full recovery model to the Simple recovery mode, and then back, SQL Server will only keep the “active” portion of the log, which is very small. Contributed by Tom Kitta.
If you need to delete all the rows in a table, don’t use DELETE to delete them, as the DELETE statement is a logged operation and can take a significant amount of time, especially if the table is large. To perform the same task much faster, use the TRUNCATE TABLE instead, which is not a logged operation. Besides deleting all of the records in a table, this command will also reset the seed of any IDENTITY column back to its original value.
After you have run the TRUNCATE TABLE statement, it is important then to manually update the statistics on this table using UPDATE STATISTICS. This is because running TRUNCATE TABLE will not reset the statistics for the table, which means that as you add data to the table, the statistics for that table will be incorrect for a time period. Of course, if you wait long enough, and if you have Auto Update Statistics turned on for the database, then the statistics will eventually catch up with themselves. But this may not happen quickly, resulting in slowly performing queries because the Query Optimizer is using out-of-date statistics.
If you use TRUNCATE TABLE instead of DELETE to remove all of the rows of a table, TRUNCATE TABLE will not work when there are Foreign Key references present for that table. A workaround is to DROP the constraints before firing the TRUNCATE. Here’s a generic script that will drop all existing Foreign Key constraints on a specific table:
CREATE TABLE dropping_constraints
INSERT INTO dropping_constraints
‘ALTER TABLE [' +
'] DROP CONSTRAINT ‘ +
t1.CONSTRAINT_NAME = t2.CONSTRAINT_NAME
DECLARE @stmt VARCHAR(8000)
DECLARE @rowcnt INT
SELECT TOP 1 @stmt=cmd FROM dropping_constraints
SET @stmt = ‘DELETE FROM dropping_constraints WHERE cmd =’+ QUOTENAME(@stmt,””)
SELECT TOP 1 @stmt=cmd FROM dropping_constraints
DROP TABLE dropping_constraints
The above code can also be extended to drop all FK constraints in the current database. To achieve this, just comment the WHERE clause.
Don’t run a screensaver on your production SQL Servers, it can unnecessarily use CPU cycles that should be going to your application. The only exception to this is the “blank screen” screensaver, which is OK to use.
Don’t run SQL Server on the same physical server that you are running Terminal Services, or Citrix software. Both Terminal Services and Citrix are huge resource hogs, and will significantly affect the performance of SQL Server. Running the administrative version of Terminal Services on a SQL Server physical server, on the other hand, is OK, and a good idea from a convenience point of view. As is mentioned in other parts of this website, ideally, SQL Server should run on a dedicated physical server. But if you have to share a SQL Server with another application, make sure it is not Terminal Services or Citrix.
Use sp_who or sp_who2 (sp_who2 is not documented in the SQL Server Books Online, but offers more details than sp_who) to provide locking and performance-related information about current connections to SQL Server. Sometimes, when SQL Server is very busy, you can’t use Enterprise Manager or Management Studio to view current connection activity via the GUI, but you can always use these two commands from Query Analyzer or Management Studio, even when SQL Server is very busy.
SQL Server uses its own internal thread scheduler (called the UMS) when running in either native thread mode or in fiber mode. By examining the UMS’s Scheduler Queue Length, you can help determine whether or not that the CPU or CPUs on the server are presenting a bottleneck.
This is similar to checking the System Object: Processor Queue Length in Performance Monitor. If you are not familiar with this counter, what this counter tells you is how many threads are waiting to be executed on the server. Generally, if there are more than two threads waiting to be executed on a server, then that server can be assumed to have a CPU bottleneck.
The advantage of using the UMS’s Schedule Queue Length over the System Object: Processor Queue Length is that it focuses strictly on SQL Server threads, not all of the threads running on a server.
To view what is going on inside the UMS, you can run the following undocumented command:
For every CPU in your server, you will get Scheduler. Each Scheduler will be identified with a number, starting with 0. So if you have four CPUs in your server, there will be four Schedulers listed after running the above command, Schedulers ID 0 through 3.
The “num users” tells you the number of SQL threads there are for a specific scheduler.
The “num runnable,” or better known as the “Scheduler Queue Length,” is the key indicator to watch. Generally, this number will be 0, which indicates that there are no SQL Server threads waiting to run. If this number is 2 or more, this indicates a possible CPU bottleneck on the server. Keep in mind that the values presented by this command are point data, which means that the values are only accurate for the split second when they were captured, and will be always changing. But if you run this command when the server is very busy, the results should be indicative of what is going on at that time. You may want to run this command multiple time to see what is going on over time.
The “num workers” refers to the actual number of worker threads there are in the thread pool.
The “idle workers” refers to the number of idle worker threads.
The “cntxt switches” refers to the number of context switches between runnable threads.
The “cntxt switches(idle)” refers to the number of context switches to “idle” threads.
As you can see, this command is for advanced users, and is just one of many tools that can be used to see internally how SQL Server is performing.