Wednesday, August 20, 2008


Determining Poorly Performing Queries for Tuning from SQL Server Workload Trace Files

Whenever you gather workload traces to identify poorly performing queries, you need to import this data into a database table, and to "normalise" and aggregate this information to identify the worst offenders. This can be done in a variety of ways. One way is to define a regular expression such as this SQL CLR method based on work done by Itzik Ben-Gan and modified by Adam Machanic:

[Microsoft.SqlServer.Server.SqlFunction(IsDeterministic = true)]
public static SqlString sqlsig(SqlString querystring)
    return (SqlString)Regex.Replace(
       @"([\s,(=<>!](?![^\]]+[\]]))(?:(?:(?:(?:(?# expression coming
       )(?:([N])?(')(?:[^']'')*('))(?# character
       )(?:0x[\da-fA-F]*)(?# binary
       )(?:[-+]?(?:(?:[\d]*\.[\d]*[\d]+)(?# precise number
       )(?:[eE]?[\d]*)))(?# imprecise number
       )(?:[~]?[-+]?(?:[\d]+))(?# integer
       )(?:[nN][uU][lL][lL])(?# null
       ))(?:[\s]?[\+\-\*\/\%\&\\^][\s]?)?)+(?# operators

Recently I’ve been trying out ClearTrace, a free tool based around Read80Trace (described and downloadable here). Read80Trace was originally part of a Microsoft PSS engineer’s internal toolkit, but was released to the public in Dec 2007 (RML Utilities for SQL Server). ClearTrace is extremely simple to use, imports files (including rollover) very quickly and the results are good. The project is being supported so if you find a SQL statement that isn’t normalised/parameterised correctly, you can click a button and report it.

The larger RML Utilities toolkit for Microsoft SQL Server was released here.

The RML Utilities can help you answer the following questions:

  • Which application, database or login is consuming the most resources, and which queries are responsible for that.
  • Whether there were any plan changes for a batch during the time when the trace was captured and how each of those plans performed.
  • What queries are running slower in today's data as compared to a previous set of data.

You can also test how the system will behave with some change (different service pack or hot fix build, changing a stored procedure or function, modifying or adding indexes, and so forth) by using the provided tools to replay the trace files against another instance of SQL Server. If you capture trace during this replay you can use the tools to directly compare to the original baseline capture.

If you decide to install and experiment with the RML Utilities toolkit, be warned that the tools are provided as is, and the install process is neither easy nor particularly pleasant!


Powered by Blogger