Monitoring SQL Server 2005 Performance : Using Windows System Monitor & Using SQL Server Profiler

Using Windows System Monitor

Microsoft Windows server operating systems offer the System Monitor graphical tool (formerly known as Performance Monitor [PerfMon]) to measure the performance of all aspects of the server. You can use System Monitor to view SQL Server and other applications from a coarse perspective, looking at events from a high level. You can view SQL Server objects and performance counters by using the System Monitor tool. You can also use it to gather information pertaining to the behavior of processors, memory, cache, and other object resources on the server.

Each object you monitor by using System Monitor has an associated set of counters. These counters are put in place when software is installed in the Windows environment. You can use them to measure aspects of an object, such as device usage, queue lengths, delays, and other factors. In general, when looking at information on how the server as a whole is operating and how application are performing in respect to the server’s performance, the System Monitor provides you with optimum benefits.

When you install SQL Server and its related components, several sets of counters are placed in the environment and made available for monitoring. The general list of objects available to be monitored depends on which SQL components have been installed. The following prefixes identify the objects you can monitor in SQL Server:

• .NET CLR— Use this prefix to access the common language runtime.

• MSAS 2005— Use this prefix to access Microsoft Analysis Services.

• MSFTESQL— Use this prefix to access the Microsoft Full-Text engine.

• MSRS 2005— Use this prefix to access Microsoft Reporting Services.

• MSSQL$InstanceName— Use this prefix to access a named instance of the database engine.

• SQLAgent$InstanceName— Use this prefix to access a named instance of the SQL Server Agent.

• SQLAgent— Use this prefix to access the default instance of the SQL Server Agent.

• SQLServer— Use this prefix to access the default instance of the database engine.

Each of the software categories has several related objects, and within each object is a set of one or more counters. Depending on the information you want to gather, you need to select the objects in the software prefix you are looking for and then select one or more counters from that object’s list. Each of the software categories has some counters that may be of interest.

As you begin any of these processes, you need to begin with the hardware and operating system and then proceed into the application server. As you get further into data gathering and analysis, you should look into each database and the interactions between the data and the user applications. To view SQL Server tasks in detail, after the initial data-gathering processes, you use the SQL Server Profiler to develop a more complete picture.

Using SQL Server Profiler

The SQL Profiler tool is a graphical mechanism that enables you to monitor SQL Server events. This tool enables you to capture and save data about every event on a server. The data can be stored to a file or SQL Server table. You can later analyze stored data replay events. This allows for the scheduling of the Profiler or other execution for deferred analysis, when time permits.

You should use SQL Profiler to monitor only the events of interest. Monitoring all events on the server produces so much data that it can become overwhelming. You can filter large traces so that you view only the information you want to see, such as a single database or single user. You can use filters to view a subset of the event data captured. Monitoring too many events also adds excessive overhead to the server. This overhead can slow the monitoring process and cause the large amounts of output from the trace to create a file or table that is large. This is particularly important when you are performing monitoring over long periods.

It is a good idea not to run the SQL Server Profiler on the same server you are monitoring. Running the Profiler uses considerable resources, and that can noticeably affect the server’s performance. Instead, you should run Profiler on another server or workstation that can act as a monitoring machine and have all data collected there.

SQL Profiler is a useful tool that you can use to do the following:

• Monitor the performance of SQL Server.

• Debug T-SQL statements and stored procedures.

• Identify long-running queries.

• Step through procedures to ensure that they are working as expected.

• Capture events on a production system and replay them on a test system.

• Diagnose problem situations by capturing event data.

• Audit and review events.

In troubleshooting the SQL Server environment, you typically use SQL Profiler. You can best use this tool to find queries that are not performing well or ones that are executing for long periods. Profiler is also useful in identifying causes of data blocking and deadlock situations. In monitoring a healthy server, you generally use SQL Profiler to monitor performance and to audit application, user, database, and job activity.

Before you start using Profiler, you should become familiar with profile templates. A template defines the criteria for each event you want to monitor with SQL Profiler. You can use predefined templates for individual circumstances, and you can create your own templates as well, specifying the events, data columns, and filters to use. You can save a template, and at any time, you can load a trace and then start it with the template settings. To help identify long-running queries, you use Profiler’s Create Trace Wizard to run the TSQL_Duration template. You can specify the length of the long-running queries you are trying to identify and then have them recorded in a log.

SQL Profiler captures data by using a trace based on the selected events, data columns, and filters. The trace is the basis for all data collected; you can define a trace on an ad hoc basis, draw a trace from a template, or a combination of the two. Even after you have defined the data collected, you can apply filters to the data after it is captured to focus on the type of information you want. You might therefore want to save traces even after you are finished with the current activity. You can possibly apply a past trace to various circumstances.

At times, when monitoring with Profiler, you may find the amount of data provided to be considerable and possibly overwhelming. It can be difficult to find what you are looking for in a trace that covers a broad range of events. A useful technique that can ease this process is to write a trace to a SQL Server table and then query the table from within the Query Analyzer. If you know what you are looking for, this method can greatly speed up finding the data in the trace you need.

Profiler can store captured data in a text file or in a table. If you decide to store the data in a SQL Server table, you should not store it in a database you are profiling, and, if possible, not even on the same server because it could affect the performance of the server you are profiling. Instead, you should store the trace data in a database on another server. After data is captured (which is the easy part of the process), you must sift through the data collected to draw some meaning from the results.

Profiler can use many existing templates to gather information for various types of circumstances. You might want to select some of them to see what information is being gathered. After a trace is created, it is permanently stored until it is deleted. You can start a trace again by name through the Profiler interface or via a stored procedure.

To use the output generated by a trace, you must first determine what type of data you are most interested in from the trace. The next section describes how to get the most valuable information from the trace results.

Defining a Profiler Trace

When using SQL Profiler to define a trace, you use event categories to select the events to monitor. Event categories are grouped into classes of events, and you can select the following classes:

• Broker— Finds events produced by the Service Broker.

• CLR— Finds events produced through the loading of assemblies.

• Cursors— Finds events produced by the use of cursors.

• Database— Finds events produced when files grow or shrink automatically.

• Deprecation— Finds events produced by elements of SQL Server that are to be removed in a future release.

• Errors and Warnings— Finds events produced when an error or warning occurs.

• Full Text— Finds events produced by use of full-text indexing and full-text queries.

• Locks— Find events produced when a lock is acquired or other lock activity occurs.

• OLEDB— Finds events produced through calls from an OLE DB interface.

• Objects— Finds events that occur as objects are created, opened, closed, or deleted.

• Performance— Finds events produced when SQL data manipulations execute.

• Progress Report —Finds events produced through online index operations.

• Scans— Finds events produced when tables and indexes are scanned.

• Server— Finds general server events produced.

• Security Audit— Finds events used to audit server activity.

• Sessions— Finds events produced by clients connecting and disconnecting.

• Stored Procedures— Finds events produced by the execution of procedures.

• Transactions— Finds events produced by the execution of Microsoft Distributed Transaction Coordinator transactions or through writing to the transaction log.

• TSQL— Finds events produced by the execution of T-SQL statements.

• User Configurable— Finds user-configurable events.

Each event class can monitor various objects. To select any of the objects when defining a trace, you use the Events tab of the Trace Properties dialog box. You add and remove objects, not whole classes, although you can trace a whole class of objects, if desired.

Using Profiler Traces to Diagnose Locking

SQL Profiler provides the Locks event classes to monitor locking behavior during trace operations. Several of these classes are useful in monitoring locking, blocking, and deadlocking situations on a server:

• Lock:Acquired— This event fires to show the acquisition of a resource lock.

• Lock:Cancel— With this class, an event is fired when a lock on a resource has been cancelled. SQL Server can cancel a lock because of a deadlock or through a programmatic process cancellation.

• Lock:Deadlock— With this class, a deadlock occurs if two concurrent transactions have deadlocked each other by trying to obtain locks on resources that the other owns.

• Lock:Deadlock Chain— With this class, you can see the chain of events produced for each of the processes leading up to a deadlock situation.

• Lock:Escalation— With this class, an event fires when the server determines that a lock should be converted to a larger scope.

• Lock:Released —With this class, an event fires when a resource lock is released.

• Lock:Timeout— This event fires when a lock request has timed out because another process is blocking a resource with its own lock.

You use Lock:Acquired and Lock:Released to monitor the timing of lock behavior. These events indicate the type of lock and the length of time the lock was held. Often, a redesign of the application that is setting the locks in place can lessen the lock duration considerably.

You use Lock:Deadlock, Lock:Deadlock Chain, and Lock:Timeout to monitor deadlock and timeout situations. This information is useful in determining whether deadlocks and timeouts are affecting the user or application.

The SQL Profiler Results window is segmented into two view panes. If you have included TextData as one of the columns in your definition, the bottom pane shows SQL statement information. The top pane illustrates the current trace data view where event information is displayed, based on current filter settings.

After you have collected trace event data, you can save the trace to replay it later. The SQL Profiler Playback feature is powerful but carries a little overhead. It is well worth considering having a test machine available to act as a playback and troubleshooting server.

Playback of events is accomplished through the SQL Server multithreaded playback engine. This engine can simulate user connections and SQL Server authentication. The event data can be played back to reproduce the activity captured in the trace. Replay can be useful in troubleshooting an application or another process problem.

Trace Playback and Diagnosis

After you have identified a problem and implemented corrections, you should run the trace that was originally collected against the corrected application or process to see whether the proposed solution accomplishes the desired effect.

Replaying the original trace can be useful in designing solutions. The trace replay feature has advanced debugging support, enabling you to make use of break points and run-to-cursor features.

When the target computer is going to be a computer other than the computer originally traced, you must ensure that the database IDs on the target are the same as those on the source. You can accomplish this by creating (from the source) a backup of the master database as well as any user databases referenced in the trace and restoring them on the target. In this manner, you can use a test server as a debugging server for any multiple-application environment.

The default database for each login contained in a trace must be set on the target. The default database of the trace activity login must be set to the database that matches that login name, even in cases in which the database name might be different. To set the default database of the login, you use the sp_defaultdb system stored procedure.

You have the option of replaying the events in the order in which they were traced. This option enables debugging, which means you can implement debugging techniques such as stepping through the trace. You can replay the events using multiple threads to optimize performance; however, this disables debugging. The default option is to display the results of the replay. If the trace you want to replay is a large capture, you might want to disable this option to save disk space.

Using Profiler to Gather a Workload

One of the most common uses of Profiler is to collect an event’s workload sampling for tuning the physical database. You can analyze the database design from a workload file by using the DTA. SQL Server Profiler provides a predefined Tuning template that gathers the appropriate T-SQL events in the trace output so it can be used as a workload for the DTA.

The DTA can use a workload file generated by Profiler to analyze performance effects. After analyzing the effects of a workload on a databases, DTA provides recommendations to add, remove, or modify physical design structures. These physical performance structures include clustered indexes, nonclustered indexes, indexed views, and partitioning.