MAXDOP – SQL Server’s Way To Avoid Parallelism

Facebooktwittergoogle_plusredditpinterestlinkedinmail

SQL Server comes with built-in support for using parallelism for query execution. That means that SQL Server will split up the execution of a query among different CPUs if it thinks that it can get the query results faster that way. The default setting in SQL Server will use the number of CPU cores on the system as the maximum number of parallel processors to use. Okay… this is great, right?

 
The Grass Isn’t Always Greener
Although the thought of running 1 SQL query but using multiple processors sounds great, sometimes it can actually sometimes cause slower queries. A couple issues I have ran in to in our production environment are:

  • Sometimes the parallelism can take away CPU from other processes that are trying to run at the same time.
  • Sometimes the time that it takes to reassemble the results is longer than the not using parallelism. What this means is that when a query is split among different processors, it will need to merge them back together before it sends it to you. Sometimes this can be a longer task then just using 1 processor the whole time.

 
MAXDOP
MAXDOP is a query hint that can be added to the end of your SQL query to control the amount of parallelism that happens. When you use MAXDOP, you will specify the maximum number of parallelism threads that SQL Server uses for that query. In fact, the word MAXDOP stand for “Maximum Degree Of Parallelism”.

Here is an example of how to use the MAXDOP query hint to limit the number of parallel threads to 1 (essentially disabling the parallelism):

To show how this affects the query, let me show you the execution plan when I supply the MAXDOP 1.
MAXDOP No Parallelism

Here is a screenshot of the execution plan of that exact same query without specifying the MAXDOP. You can see that it now uses parallelism.

MAXDOP Parallelism

 
Will Removing MAXDOP Help My Queries Run Faster
Is MAXDOP a silver bullet for making queries faster? No. In most cases you will just want SQL Server to manage the parallelism. This is usually the fastest/best way for your query to execute. But occasionally if you are having a slow query, you can try changing the MAXDOP setting. You will need to test your query before and after to see if there is significant improvement.

 
How Do You Know If Your Query Is Using Parallelism?
Here are a couple ways that I use to find out if parallelism is happening to a query.

  1. Check out the actual execution plan. After you run the query and the execution plan shows up, you will see a “Parallelism” step. You can see the screenshot above for what that looks like.
  2. Monitor the Activity Monitor in SQL Server Management Studio. This is a great way to detect if your production applications are causing parallelism. It’s really easy to see in Activity Monitor because it looks like the same line repeated many times.
    MAXDOP Activity Monitor

 
Reference: http://msdn.microsoft.com/en-us/library/ms181714.aspx

SQL Server 2014 Feature – Non-Clustered Indexes For Table Variables

Facebooktwittergoogle_plusredditpinterestlinkedinmail

This is truly an amazing feature to add to SQL Server 2014. I use table variables all the time and this was the only thing I didn’t like about them. Up to this point, SQL Server did not support having non-clustered indexes on table variables (the one with the name that starts with the @). If you wanted to do this, you had to create/use a temp table (the one with the name that starts with #).

With SQL Server 2014, they changed this and now allow the table variables to have non-clustered indexes as well. To do this, we simply add a little bit of extra code after the column declaration.

In the sample above, we create a table variable called @AnimalTableVar. Then when we declare the column AnimalName, we add an index called IX_AnimalTempTable_AnimalName.

Here is how selecting from this table shows up in the execution plan!
Non-Clustered Indexes For Table Variables

Reference: http://msdn.microsoft.com/en-us/library/ms188927(v=sql.120).aspx

FORCE ORDER Query Hint

Facebooktwittergoogle_plusredditpinterestlinkedinmail

When querying out data, SQL Server has a ton of different query hints. One less commonly known query hint is FORCE ORDER. So here we will talk about what this query hint does and how we can use it efficiently.

FORCE ORDER
When you put this query hint on to your query, it tells SQL Server that when it executes the statement to not change the order of the joins in the query. It will join the tables in the exact order that is specified in the query. Normally the SQL Server optimizer will rearrange your joins to be in the order that it thinks will be optimal for your query to execute.

To use the FORCE ORDER query hint, you will want to add this at the end of your query.

 
Example Without FORCE ORDER

This is the actual execution plan without the FORCE ORDER hint. You can see that SQL Server starts at the table Animal and AnimalHabitat tables, then goes to the Habitat table. If you look at our query above, we clearly list the tables in the order of Habitat, AnimalHabitat, then Animal.

Execution Plan Without Force Order

 
Example With FORCE ORDER

In the screenshot below, it shows the actual execution plan with the FORCE ORDER hint. In the new query, you can see that SQL Server starts with the Habitat and AnimalHabitat tables and then moves to the Animal table. It does this because when we turn on the FORCE ORDER option, it uses the order of the tables specified in the query syntax.

Execution Plan With Force Order

 
What Is This Used For
The FORCE ORDER query hint is only used when you want to override the way that SQL Server wants to execute this query. Normally you will just let SQL Server figure out how to get the data from the database. It does such a good job at it, that you do not usually need to override this functionality.

Occasionally SQL Server will not figure out the most optimal way to pull this data out. In this case you may want to try this query hint.

Difference Between UNION and UNION All

Facebooktwittergoogle_plusredditpinterestlinkedinmail

The UNION operation and the UNION ALL operation perform almost the same operation.  They are both used to combine two result sets in to one result set.  The main difference between the two operations is that the UNION operation will return the unique records in the final result set.  The UNION ALL operation will return any duplicates in the final result set.

 
In the examples below, we will assume we have two tables with the following data.  Notice that the values Cat and Dog exist in both tables.

Table1 Table2
Alligator Cat
Beaver Dog
Cat Eagle
Dog Frog

 
The UNION Operation

Result
Alligator
Beaver
Cat
Dog
Eagle
Frog

Notice that the duplicate Cat and Dog values were removed from the result.

 
The UNION ALL Operation

Result
Alligator
Beaver
Cat
Dog
Cat
Dog
Eagle
Frog

Notice that the duplicate Cat and Dog values are left in the result.

 
Speed Considerations

One final difference between UNION and UNION ALL is the speed difference.  Because the UNION operation needs to return a distinct set of values, this will take extra processing time.  A general rule is that if you know that the combined data is going to be unique already… just use the plain UNION ALL operation.  Only use the UNION operation if there are duplicates that you would like to have remove.
NOTE:  The UNION and UNION ALL operations are actually one operation.  The ALL keyword is just an attribute.  Because they are so different, I find it is easier to think of them both as separate operations.

 
Reference:  http://msdn.microsoft.com/en-us/library/ms180026.aspx
 

SQL Server 2014 Feature – Timeout For Online Index Rebuilding

Facebooktwittergoogle_plusredditpinterestlinkedinmail

SQL Server 2014 comes with a new argument for how to handle the blocking issues that come from online index rebuilds.

 
Before SQL Server 2014

Before the existence of SQL Server 2014, if you were doing an online index rebuild… SQL Server would wait until the index rebuild could get the locks that it needs for rebuilding the index.  This can cause long running maintenance plans on very active tables.

 
SQL Server 2014 Online Index Rebuild Argument

The new functionality available in SQL Server 2014 tags on to the original index/table rebuild syntax.  It allows you to customize how you want to handle the blocking that occurs on the indexes/tables that are being rebuilt.  Here is a sample of the new syntax for managing the blocking on the index rebuild.

Notice in the statement above, the new WAIT_AT_LOW_PRIORITY flag that gets added to the ONLINE argument.  When you rebuild an index in SQL Server with the WAIT_AT_LOW_PRIORITY flag turned on, it will allow other operations to proceed while the index build waits for low priority locks.  This argument comes with 2 parameters.

 
MAX_DURATION

The MAX_DURATION parameter is how long you want to wait in minutes.  It must always be in minutes.

 
ABORT_AFTER_WAIT

This parameter comes with 3 options.  This is basically who loses if the rebuild process cannot get the locks that it needs.  The connection that you choose will be terminated after the timeout period.

  • NONE – Continue waiting for the lock
  • SELF – The rebuild operation will fail
  • BLOCKERS – All transactions that are blocking the rebuild will be killed

 
If you do not specify WAIT_AT_LOW_PRIORITY with you index rebuild, then SQL Server will automatically set it to the default of 0 minutes and NONE.

 
 
Index Reference:  http://msdn.microsoft.com/en-us/library/ms188388.aspx

Table Reference:  http://msdn.microsoft.com/en-us/library/ms190273.aspx