SQL Server 2008 Management Data Warehouse
by Kalen Delaney, a SQL Server MVP since 1993 who has been working with SQL Server for over 20 years. Her major areas of specialization are internals and tuning. Find out more about Kalen’s activities through her web site at www.SQLServerInternals.com.
SQL Server 2008 introduces the Management Data Warehouse (MDW) to SQL Server Management Studio for streamlined performance troubleshooting. MDW is a set of components that enable a database developer or administrator to quickly track down problems that could be causing performance degradation. In this paper, we’ll look at the following components:
- The Data Collector
- The MDW database
- The MDW reports
Many of the reports included in MDW will provide suggestions for improving performance, based on information gathered by the Data Collector in the following areas:
- Query performance statistics and use of indexes
- Resource locking and blocking
- Disk usage
Sometimes the reports can lead directly to a solution to a performance problem. For example, if the query performance statistics report shows an extremely slow running query, you can drill down to the query plan, which may indicate a missing index. In many cases, creating that index can make a major difference in the query performance. In another case, the query performance statistics report might show the same query performing sporadically, sometimes very fast and sometimes slow. You might then determine that the slower executions are using a different execution plan. This could lead you to capturing the plan for the fast running execution and creating a plan guide. A third example might be that the resource locking and blocking report indicates unacceptable lock wait times when running queries in a particular database. This might suggest that you explore the possibility of using one of the snapshot based isolation levels, to alleviate the blocking problems.
Understanding exactly how to interpret all of the details in the reports, and knowing how to correct the poor performance shown in the reports requires knowledge of SQL Server that is beyond the scope of this paper.
Overview of Management Data Warehouse
Setting up the MDW is a two-step process. First, you must run the “Configure Management Data Warehouse” Wizard to create the MDW database. The second step is to set up data collection by specifying the database that you want to upload the data to. The wizard then configures the data collection using the SQL Server Integration Services runtime. This second step requires that you run the “Configure Management Data Warehouse” Wizard a second time. SQL Server Agent jobs are created to start and stop the packages that perform the data collection and the uploading of the data. It is strongly recommended that you not modify the jobs directly. Although the jobs behave like any other SQL Server Agent jobs, the function is only to control the execution of the SSIS package.
Although the purpose of the paper is not to provide an installation tutorial, it will be useful to refer to Figure 1, which is the first dialogue of the “Configure Management Data Warehouse” Wizard. The wizard is accessed from SQL Server Management Studio by right-clicking on the “Data Collection” icon in the Management folder. When you select the option to “Create or upgrade a management data warehouse” and click “Next”, you are given a dialog to create a database in which to store the collected performance data. The default database name is MDW, and it is recommended that you use that name as most of the documentation assumes that is the name of the database containing the collected data. In particular, you should not change the database name after creation, because all of the jobs created to manage the database collection refer to the database by the original name and will generate errors if the name is changed. However, you can create a new database, and reconfigure the data collector to store its collected data to use it by rerunning the “Configure Management Data Warehouse” wizard again.
Figure 1: Starting the MDW configuration process
After creating the database, the wizard will ask you to map logins and users to management data warehouse administrator, reader and writer roles. The default is not to allow anyone but the database owner to have access to the MDW database.
Best practice tip: You should not change the database name after creation, because all of the jobs created to manage the database collection refer to the database by the original name and will generate errors if the name is changed.
After creating the MDW database, you will need to run the “Configure Management Data Warehouse” Wizard a second time to configure the data collection and create the jobs. The wizard will also ask you for a directory for caching collected data prior to uploading to the MDW database. (Caching will be discussed in more detail in the next section, “Collection”). If you leave the directory specification blank, the cache will be in the TEMP directory for the Service Account that your SQL Server Agent Service is using. (If your SQL Server Agent service is running under LocalSystem, the cache will be in the TEMP directory for the user that is running the wizard. For example, my cache is stored in “C:\Users\Kalen\AppData\Local\Temp”. When choosing a location for the cache, you should pick one that will be least disruptive to the target SQL Server instance when lots of writes are being performed. You also need to make sure that the SQL Server Agent account has privileges to the location. You also need to make sure that the SQL Server Agent account will have access to the MDW database if it has been configured on a remote server. To ensure this access, you can create a SQL Server Agent proxy account and add it to the dc_proxy database role for the msdb database.
Figure 2 shows the last screen of the wizard, indicating the various steps the “Configure Management Data Warehouse” Wizard will perform this time.
Figure 2: Configure Management Data Warehouse Wizard, final step
The Data Collection is performed primarily through SSIS packages that control the collection frequency on the target and performs the upload of data to the MDW. The SQL Agent jobs control the triggering of the upload action for SSIS.
Figure 3 shows the existing jobs after completing the configuration. The first five jobs seen in the figure are the jobs that trigger the data collection and upload packages. The sixth job (mdw_purge_data_[MDW]) was created when the MDW database was created, and the last job shown is associated with Policy Based Management (PBM) and is not in the scope of this paper.
Figure 3: Existing jobs in SQL Server Agent after configuration
The next section on “Data Collection Architecture” provides more details on collection and upload. For now, you should be aware of the following recommendations:
Do not change any of the job specifications for the data collection and upload jobs.
You can change properties of the mdw_purge_data_[MDW] job, which will remove data from the MDW database on a regular basis, subject to a retention policy. This job will also purge the job history from the data collection and upload jobs.
In general, the retention setting is controlled via property pages for specific collection sets. The mdw_purge_data job triggers when the stored procedure that performs the purge operation starts. The only property that you should change is when to perform the purge operation along with any alerting that you would like to add. You should not modify the job because any changes that you might add would get overwritten with subsequent software upgrades of SQL Server. If you want to perform archiving before the mdw_purge_data job runs, you can create a separate job that runs prior to the mdw_purge_data job.
Best practice tip: You should not change any of the job specifications for the data collection and upload jobs within SQL Server Agent.
To understand the data collection processes, the following terms will be useful:
A target is an instance of SQL Server from which data is being collected. Note that SQL Server Express instances cannot be targets.
- Target Type
A target type is an aspect of the target from which data is being collected. For example, a target type could be an entire SQL Server instance, or a single database in that instance. Each target type has different characteristics and different performance data that can be collected.
- Data Provider
A data provider is a data source, which is specific to a target type. For example, system views can be data providers. Most system views provide data for a SQL Server instance target, but some provide data for a SQL Server database target.
- Collector Type
A collector type is a logical wrapper around an SSIS package that provides the actual mechanism for collecting and uploading data to the MDW database. Collector types include: TSQL Query collector type, SQL Trace Collector type, Performance Counter collector type and query activity collector type
- Collection Item
A collection item is data gathered from a single data provider, such as a specific query against a system view, or the value of a specific performance monitor counter. A collection item is an instance of a collector type.
- Collection Set
A collection set is a group of related collection items. There are three collection sets defined when you configure your MDW: one that contains items pertaining to disk usage, one containing items pertaining to query execution and performance statistics, and one containing items pertaining to server activity, such as CPU and memory usage, as well as waittype and blocking data.
- Collection Mode
The collection mode is the manner in which data is collected and stored.
After running the “Configure Management Data Warehouse” Wizard, you will have three default collection sets in two collection modes. The Disk Usage collection set is in non-cached mode and the Server Activity and Query Statistics collection sets are in cached mode. The caching mode determines the relationship between the collection of data and the uploading of data to the MDW database.
In non-cached mode, collection and upload are on the same schedule. The packages start, collect, and upload data at their configured frequency, and run until they are finished. There is only one job created for collection sets in non-cached mode. Non-cached mode allows for the collection and uploading of data to occur on demand, as well as at the intervals specified in the job.
- Cached Mode
In cached mode, the data collection and uploading of jobs are on different schedules, so two jobs are created for each collection set. Cached mode supports continuous collection of data, but less frequent uploads. For example, the default Server Activity collection set uploads data by default every 15 minutes. However, the collection items are collected every 60 seconds.
Table 1 highlights the key features of each of the three default collection sets. You can see the mode (cached or non-cached), the default upload frequency, the default data retention period, and a summary of the collection items in each of the collection sets.
Table 1 Summary of the default collection sets
Description of Technology
The data collection process consists of scheduled jobs to collect and upload performance data into the configured database.
Data Collector Architecture
Figure 4, found in SQL Server Books Online, shows the components that make up the data collection process. The collection system is grouped into four categories:
- Storage components are the databases that contain configuration information and jobs (msdb) as well as the collected performance data (MDW).
- Execution components manage the data collection and storage operations.
- API components enable interaction between the user interfaces and the data collector and the user interfaces. These components include SQL Server metadata, such as the dynamic management views.
- Client components are the user interfaces for managing the data collection and presenting the reports.
Figure 4: Components of MDW Data Collection
The default configuration creates 5 jobs to manage the three default collection sets. In addition, a job to purge data, both from the MDW database and from the job history, is created. The names of five data collection jobs all start with collection_set. The Disk Usage Collection set only has a single job, since it is non-cached, and the name of the job reflects this: collection_set_1_noncached_collect_and_upload. The Query Statistics collection set has two jobs: collection_set_2_collection and collection_set_2_upload. Server Activity also has two jobs: collection_set_3_collection and collection_set_3_upload.
Scheduling the Collections
Both of the upload jobs are scheduled to run every 15 minutes starting at midnight. For a single instance (or even just a few instances), this schedule should be fine. If you have many servers trying to upload data to the MDW database at the same time, you can consider staggering the start times so that the uploads are not running concurrently. For example, you could have the collection_set_2_upload job start at 12:01 and the collection_set_3_upload job start at 12:06.
Best practice tip: Consider staggering start times for upload jobs so they are not running concurrently to help avoid a tax on performance.
Although the storage component comprises both the msdb database, where jobs, history, SSIS packages, and configuration information are stored, as well as the MDW database, we will only be looking at the MDW database in this section. If you have worked with scheduling jobs using SQL Server Agent, the jobs and history component of MDW should not present any special challenges. If you have worked with SSIS, the packages for managing the data collection and uploads should look like normal jobs.
The MDW database contains the data collected by the default collection sets and any custom collection sets you define. There are two schemas in the database: core and snapshots. The core schema contains the ‘dimension’ tables that may be relatively small. For example, one of the tables in the core schema is core.wait_categories that is used by the Server Activity collection set and contains only 17 rows, one for each category of wait, including Backup, Latch, Lock, Logging, etc. The tables in the snapshots schema are the ‘fact’ tables that contain the actual collected data. These tables can contain millions or billions of rows, depending on how busy your SQL Server is and how often your purge jobs run.
There is one additional aspect of storage that does not show up in Figure 4, and that is the cache location, where the data from the jobs running in cached mode is stored prior to being uploaded to the MDW database. Although the setup process will use a default value for this directory, it is recommended that you carefully consider the placement of this caching directory so as not to interfere with the disk activity writing to the MDW database itself.
Configuration of Collections
As mentioned, after the jobs are created, you should not change any of their properties as they are used mainly to trigger the SSIS packages. However, you can configure some of the properties of the of the collection sets, by going to the data Collection node in the Management folder of SQL Server Management Studio. Right click on any of the collection sets, and you can control the upload schedule, whether the job runs as cached or non-cached, and the data retention period. You can also disable the entire data collection process by right clicking on the Data Collection node and selecting “Disable Data Collection”. Doing this will stop all the data collection and upload jobs. Individual collection sets can be stopped, and this would usually be done to support a specific troubleshooting scenario. You may only want to start a specific collection set for troubleshooting a known problem or test scenario.
Metadata to Support Collections
SQL Server 2008 provides system stored procedures and system metadata views to help you support your MDW collection sets.
The system procedures provided enable you to programmatically create and manage the data collector and data collection. There are also procedures to allow you to create your own custom collector sets. The procedures fall into the following categories:
- Data Collector stored procedures
- Collection sets stored procedures
- Collection items stored procedures
- Collector types stored procedures
- MDW configuration stored procedures
- Package upload stored procedures
- Data collection execution log stored procedures
Best practice tip: User custom collections for collecting more queries than the default setting; the top 3 worst performing queries.
Views and Functions
The views and functions allow you to programmatically retrieve information about your data collector and data collection. The views provide about collection sets, collection items, and execution. The functions provide SSIS package execution and SQL Trace information.
For complete details of all the procedures, view and functions, please refer to Books Online. Most of the operations invoked by the system procedures and most of the information available through the views and functions are also available through the graphical interface. The main exception are the mechanisms to create your own data collection sets, data collector types, and data provider types, which can only be created using Transact-SQL. Describing the creation of your own collector sets, collector types, and provider types is beyond the scope of this paper.
Custom Collection Sets
Creating a custom collection set is relatively straightforward, using the provided stored procedures. To create a collection set based on performance counters, DMVs, or TSQL queries, you should take a look at the definition of the default Server Activity collection set. You can easily see this definition by right clicking on Server Activity (in the System Data Collection Sets folder in SQL Server Management Studio) and selecting “Script Data Collection”.
There is also GUI support for creating a collection set based on SQL Trace data. Once you have defined your trace using the SQL Server Profiler, select File > Export > Script Trace Definition > For SQL Trace Collection Set. That will generate a script that creates a custom trace data collector set based on a trace definition.
One of the reasons you might create a custom trace data collection set is that the default system data collector for Query Statistics does not store all of the statements – only the worst performing according to the algorithms chosen in the collection set, as documented in the Books Online. You may want to collect more queries than the top 3 worst performing ones, and if so, you could disable the default system collector and set up your own trace collection set.
There is also a mechanism for creating your own custom data collection provider, which is beyond the scope of this paper.
Details of Data Collected
We saw an overview of the information collection through each of the default collection sets in Table 1. In this Table, we saw that the non-cached collection (Disk Usage) was uploaded every 6 hours and retained for 730 days. The cached data, which is expected to be much more voluminous, was uploaded every 15 minutes and only retained for 14 days.
Server Activity Information
The server activity information falls into four general categories: waiting processes, memory usage, CPU/scheduler usage, and disk I/O. Use the information from this collection set to help troubleshoot problems with your hardware configuration, your operating system configuration, and your SQL Server’s ability to handle the workload you are submitting to it. If there are issues with concurrent access to resources, whether data resources, memory resources, or scheduler resources, the information available can indicate where the blockages are. The server activity data is gathered by running seven different queries against dynamic management views, and by collecting performance counters across the platform.
Best practice tip: User Server Activity Information to troubleshoot hardware and operating system configurations. Using Query Statistics Information for customer collection sets is not recommended as it simply duplicates the functionality of the QS collection set.
Query Statistics Information
The query statistics information collection set contains a single collection item that uses the Query Activity collector type. This special collector type was created solely to support the query statistics collection set. It’s not externally configurable and it’s not really intended for reuse in custom collection sets.
Disk Usage Information
The disk usage information gathered in a non-cached mode provides information about the space allocated and used by each of your database files and by the transaction log. By default, this information is collected and uploaded every six hours, which should be sufficient to allow you to predict your database growth, and plan for your storage needs.
Viewing the Data
The third major component of the Management Data Warehouse is the graphical reports available to view the collected data. At a high level, you can look at the collection of reports across multiple servers, by right clicking on your MDW database, as shown in Figure 5.
Figure 5: Accessing Management Data Warehouse reports
Figure 6 shows you an example of an overview report spanning 13 separate SQL Server instances. The Management Data Warehouse Overview report displays the last upload time so that you can get a quick status of all the upload activity to the central MDW. The reports are also accessible via the right click menu on the Data Collection node for the target instance.
Figure 6: Management Data Warehouse reports, spanning SQL Server instances
Using SQL Server Management Studio Drilldown Reports
The most powerful feature of the visual reports is the ability to drill down into the details. Here are some examples.
You can click on the last Query Statistics upload time for any of the server instances in Figure 6 and you’ll see a report like that shown in Figure 7.
Figure 7: Query Statistics
The timeline along the top lets you choose the time period in which you are interested. The middle graph shows the top 10 queries for the selected time period. By default, the top queries by CPU are shown, but you can click on one of the other measures instead and see the top queries, by duration, I/O, physical reads or logical writes. The bottom section the report reports the TSQL text for the top 10 queries. You can click on any of the queries and see a drilldown report like the one in Figure 8.
Figure 8: Query Statistics Drilldown Report
If you scroll down on the screen above, you see an additional report that provides details of the query plans for the collected queries. This report gives you the ability to drill into the graphical execution plan, where you’ll be able to determine which indexes are being scanned, and when the optimizer determines that there is a missing index that could improve performance if it were created.
The Server Activity report shows you system resource usage information and SQL Server wait statistics, as shown in Figure 9.
Figure 9: Server Activity Report
Again, you can choose the time frame from the timeline at the top. The next section allows you to select one of the system resource reports and drill down into details of CPU, Memory, Disk I/O, and Network Usage. The next section shows SQL Server waits for the most common waittype categories. Clicking on one of the bars drills down into the details of wait types for that time period as shown in Figure 10.
Figure 10: Server Activity Drilldown
There are two more very useful reports that you can generate from this point. First, if you click on the Wait Category “Lock”, you’ll see the Blocking Report that indicates what processes are being blocked, what processes are causing the blocking, and what resources the blocked processes are waiting to access. This report can help you track down blocking chains, as well as help you detect which resources are causing the most contention. Second, if you click on any of the other Wait Category values, a report showing General Waits will be shown.
Disk Space Usage
You can click on the last Disk Usage upload time for any of the server instances in Figure 6, and see a report similar to Figure 11, which indicates the data file size and log size for each database on the instance at the time the last upload was made.
Figure 11: Disk Space Usage, data file size and log size
All of these reports can also be accessed from the Data Collection Node of the Management folder in the SQL Server Management Studio. Right click on Data Collection, select Reports, Management Data Warehouse, and then select one of the Collection Sets. You’ll access reports similar to Figure 7, 9 or 11.
It is also possible to create your own custom reports, using SQL Server Reporting Services and querying the MDW database directly. If you want your report to work inside of Management Studio, you’ll need to use the report designer that was released with BIDS with SQL Server 2005. This is because SQL Server Management Studio uses the client side reporting control from Reporting Services 2005.
Viewing the Collection Set Logs
There is also data available in the data collection logs. You can right click the Data Collection Node or any of the nodes under it, and select View Logs. From the dialog that opens, you can choose which of the collection sets you want to see the logs for, and also choose to see the entire available history of the jobs performing the collection and upload for that collection set.
Best Practices and Caveats for Management Data Warehouse
Here are some best practices to keep in mind:
- Use a centralized server for the MDW database. This allows you to use a single point for viewing reports for multiple instances. You’ll need to be within the same domain for this to work.
- When creating a trace collection set, be sure to set filters to collect just the data you need to diagnose the problem. For example, create a trace for a set of stored procedures that only collect data if the duration of the query exceeds a threshold. This way you don’t flood the MDW database with meaningless data.
- Before adding a collection item with performance counters, make sure you thoroughly understand what the default Server Activity collection set is already collecting, so that you do not end up collecting the same data through multiple collection sets.
The costs associated with the MDW are mainly the costs of data collection and each collection set has its own process (an instance of DCEXEC.exe) performing the collection. To minimize the overhead of the collection processes, keep the following points in mind:
- If you are collecting the results from multiple queries that have the same collection frequency, combine the queries into a single collection item. The XML parameters for a single T-SQL collection item can have multiple <Query> elements. Doing this will minimize the amount of memory that DCEXEC.exe uses during collection. Similarly, combine multiple performance counter collection items into a single collection item wherever possible.
- Combine collection items into a single collection set whenever possible. The only reason for separate collection sets is if you need separate data retention periods or different collection schedules. For example, suppose you wanted to capture the output of 6 DMV queries. You want to keep the data from the first two queries for 6 months, but it is acceptable to purge the data from the remaining queries after just two weeks. You should not create a separate collection set for each query; two collection sets will capture this data with much less overhead.
- A collection set running in cached collection mode will always have a collection process running. If you collect data very frequently, this is more efficient than starting and stopping a new process every time new data must be collected. In contrast, a collection set in non-cached mode will not have a collection process running most of the time. A new process will be started at the scheduled collection and upload time; it will collect and upload data in a single step, then exit. If you are collecting data infrequently, non-cached mode is more efficient than leaving a generally-idle process running all of the time. As a general rule of thumb, if you will be capturing data once every 5 minutes or more frequently than this, consider using a collection set that runs in cached collection mode. If it will suffice to collect data less frequently than every five minutes, consider using non-cached collection mode, instead.
- Collection frequency for cached collection sets can be as high as one execution every five seconds. You should be aware that more frequent collection has correspondingly high overhead. Always choose the lowest collection frequency that will meet your needs.
- One DCEXEC.exe instance can have a memory working set of 30 – 100 MB, depending on the items in the collection set.
- Starting a new DCEXEC.exe process for non-cached collection or for data upload can take 10 seconds or more of wall clock time, much of it CPU-bound. In most scenarios this wouldn’t be noticed, but it could be have a more measureable impact if you have very little RAM (1GB or less), if you have many collection sets, or if you don’t follow the best practices listed above.
Caveats and Limitations in the Current Version
In the current version, removing data collection once it has been configured is not supported. In addition, you should not drop or change the name of the MDW database, as all of the jobs are based on the original database name.
Summary and Conclusion
The Management Data Warehouse in SQL Server 2008 is a powerful collection of components which provide an easy way to create a repository of performance-related information. We encourage you to start exploring the data collected using MDW, even if you are not an experienced SQL Server tuner.
The Management Data Warehouse consists of three components:
- A set of custom data collectors, which are based on SQL Server Agent jobs that collect and store performance data
- A database (called MDW, by default) for holding the data repository
- A set of graphical drilldown reports for viewing the data collected
Custom collectors can be added and custom reports can be developed and displayed to capture and present the data that is important to you and your troubleshooting efforts.
You are welcome submit community comments about how you are using the data collected and what you would like to see in the product, as this can drive future development.
For more information
SQL Server 2008 Internals, Microsoft Press, 2009
About the author. Kalen Delaney**, a SQL Server MVP since 1993, has been working with SQL Server for over 20 years. Her major areas of specialization are internals and tuning. Kalen teaches her advanced SQL Server curriculum to clients around the world, including at many regional Microsoft offices. Kalen writes regularly for SQL Server Magazine and is the primary author for the Inside SQL Server series of books, for SQL Server 7.0, SQL Server 2000, and SQL Server 2005. Her latest book, from Microsoft Press, is SQL Server 2008 Internals. You can find out more about Kalen’s activities through her web site at* www.SQLServerInternals.com*.