Approach to SQL Performance issues

Performance degradation can happen due to multiple reasons. The main bottlenecks that would affect performance are:

1. Memory: Physical or Virtual – Your server could be experiencing a physical crunch which would point us to the Available Physical RAM or a crunch on the Virtual Address Space. This would involve finding out the queries which are consuming high memory during compilation or execution and narrow down on the single biggest contributor and work towards rectifying the problem with the query.

2. Disk IO: There could be an issue with SQL Server queries running slowly due to large amount of IO being done by them or a disk performance issue.

3. CPU: It could be because the queries executing on your server are consuming high amount of CPU time. In such a case, it be statistics that are out-of-date or data that has increased which introduces data skew.

4. Blocking: The queries executing on the server could be blocked and waiting for a particular resource and thus increasing the query duration.

How to perform post-mortem analysis for SQL performance problems?

If you are facing a performance degradation on the SQL instance after an upgrade, we would need to analyze the PSSDIAG collected to find out the top bottlenecks. There are two ways to do this:

1. Comparison Analysis: You identify the a query batch or stored procedure which is a problem and provide us a baseline (w.r.t. normal execution time for the query in question). We capture a “fast” and a “slow” run and we perform a comparison between the two runs and we tell you what is biggest bottleneck: Memory, Disk, CPU or Database/Environment Settings like database statistics, query SET options, difference in parameters etc. Based on what is identified as the culprit, you will be able to identify/suggest action items to help remove the culprit.

2. Bottleneck Analysis: In case you do not have a baseline, you can perform a bottleneck analysis and then stack rank them for you in case there is more than one bottleneck. Based on the analysis, you will be able to identify the biggest bottleneck and can implement action items to mitigate the single largest bottleneck.

If you need to perform live troubleshooting on the server, then you can make use of DMVs if you are on SQL Server 2005 or above or use Performance Dashboard (mentioned below).

RML Utilities Download:

http://www.microsoft.com/downloads/details.aspx?FamilyId=7EDFA95A-A32F-440F-A3A8-5160C8DBE926&displaylang=en

More Info on how to use the RML Utilities:

http://blogs.msdn.com/psssql/archive/2007/12/18/rml-utilities-for-microsoft-sql-server-released.aspx

http://sqlcat.com/technicalnotes/archive/2008/02/01/precision-performance-for-microsoft-sql-server-using-rml-utilities-9-0.aspx

If you need to analyze blocking data on the server, then you can use SQL Nexus for the same.

SQL Nexus Download Link:

http://www.codeplex.com/sqlnexus

Another option would be to use SQL Server 2005 Performance Dashboard in case you are testing your application on SQL Server 2005. This tool lets you view your server status without running PSSDIAG and provides reports to identify Long Running queries and also identify which queries are consuming the highest amount of resources (IO/CPU/Memory).

Performance Dashboard:

http://www.microsoft.com/downloads/details.aspx?FamilyId=1d3a4a0d-7e0c-4730-8204-e419218c1efc

Performance Dashboard for SQL Server 2008

http://blogs.msdn.com/b/sqlserverfaq/archive/2010/05/27/sql-server-performance-dashboard-reports-in-ssms-introduction-install-datediff-error-amp-modified-for-sql-2008.aspx

Advertisement

System.OutOfMemoryException for Management Studio

I have seen multiple posts on the web trying to decipher the OutOfMemoryException thrown by Management Studio (SSMS) while executing a user query. Contrary to popular beliefs, this is not a SQL Database Engine Out of Memory condition. This is a client side OOM condition thrown by the .NET runtime while trying to receive the output of the query from the Database Engine. More often than not, the culprit is a large number of rows returned by the query submitted by the user.

A lot of times the error can be circumvented by using the TEXT mode output of SSMS. The error is a .NET Out of Memory exception pointing to the Management Studio running out of available physical memory. The GRID output requires a .NET GridView control to be created where as the Text mode output uses a TextBox to store the output returned by the database engine. The amount of memory consumed by the GRID is higher than a TextBox.

It is always advisable to store the output of a query returns a large number of rows into a file (CTRL+SHIFT+F) or use SQLCMD to generate the output into a CSV/TXT file. This would help optimize the memory usage on the box that is executing the query and also prevent re-execution of the query due to client box out-of-memory conditions. IMHO I cannot fathom the need to output a million rows in the GRID view because it is not possible to parse the output unless you put that into a flat file! Hope this sheds some light on this common misconception.

Once such issue is mentioned below:

Reference: OOM error when we access Schema changes report from SSMS – Microsoft