This tutorial shows how to use HammerDB to perform load testing on a Google Compute Engine SQL Server instance. You can learn how to install a SQL Server instance by using the following tutorials:
There are a number of load-testing tools available. Some are free and open source, while others require licenses. HammerDB is an open source tool that generally works well to demonstrate the performance of your SQL Server database. This tutorial covers the basic steps to use HammerDB, but there are other tools available, and you should select the tools that align best to your specific workloads.
- Configuring SQL Server for load testing.
- Installing and running HammerDB.
- Collecting runtime statistics.
- Running the TPC-C load test.
In addition to any existing SQL Server instances running on Compute Engine, this tutorial uses billable components of Google Cloud Platform (GCP), including:
- Google Compute Engine
- Windows Server
The Pricing Calculator can generate a cost estimate based on your projected usage. The provided link shows the cost estimate for the products used in this tutorial, which can average 16 dollars (US) per day. New Cloud Platform users might be eligible for a free trial.
Before you begin
Sign in to your Google Account.
If you don't already have one, sign up for a new account.
Select or create a GCP project.
Make sure that billing is enabled for your project.
- If you aren't using Windows on your local machine, install a third-party RDP client such as Chrome RDP by FusionLabs.
Configuring the SQL Server instance for load testing
Before you start, you should double check that your Windows firewall rules are set up to allow traffic from the IP address of the new Windows instance you created. Then, create a new database for TPCC load testing and configure a user account using the following steps:
- Right-click the Databases folder in SQL Server Management Studio, and then choose New Database.
- Name the new database "TPCC".
- Set the initial size of the data file to 190,000 MB and the log file to 65,000 MB.
Set the Autogrowth limits to higher values by clicking the ellipsis buttons, as shown in the following screenshot:
Set the data file to grow by 64MB to unlimited size.
- Set the log file to disable auto-growth.
In the New Database dialog, in the left pane, choose the Options page.
- Set Compatibility level to SQL Server 2012 (110).
Set the Recovery model to Simple, so that the loading doesn’t fill up the transaction logs.
Click OK to create the TPCC database, which can take a few minutes to complete.
- The preconfigured SQL Server image comes with only Windows Authentication enabled, so you will need to enable mixed mode authentication within SSMS, by following this guide.
- Follow these steps to create a new SQL Server user account on your database server that has the DBOwner permission. Name the account "loaduser" and give it a secure password.
- Take note of your SQL Server internal IP address by using the Get- NetIPAddress commandlet, because it’s important for performance and security to use the internal IP.
You can run HammerDB directly on your SQL Server instance. However, for a more accurate test, create a new Windows instance and test the SQL Server instance remotely.
Creating an instance
Follow these steps to create a new Compute Engine instance:
In the Google Cloud Platform Console, go to the VM Instances page.
Set Name to hammerdb-instance.
- Set Machine type to at least half the number of CPUs as your database instance.
- In the Boot disk section, click Change to begin configuring your boot disk.
- In the OS images tab, choose Windows Server 2012 R2.
- In the Boot disk type section, select Standard persistent disk.
- Click Select.
- Click Create.
Installing the software
When it's ready, RDP to your new Windows Server instance and install the following software:
After you install HammerDB, run the
hammerdb.bat file. HammberDB does
not show up in the Start menu’s applications list. Use the following command
to run HammerDB:
Creating the connection and schema
When the application is running, the first step is to configure the connection to build the schema.
- Double-click SQL Server in the Benchmark panel.
- Choose TPC-C, an acronym that stands for:
Transaction Processing Performance Council - Benchmark C.
From the TPC.org site:
TPC-C involves a mix of five concurrent transactions of different types and complexity either executed online or queued for deferred execution. The database is comprised of nine types of tables with a wide range of record and population sizes. TPC-C is measured in transactions per minute (tpmC).
In the Benchmark panel, next to SQL Server, click the + to expand the options.
- Below TPC-C, click Schema Build and then double click Options.
Fill out the form to look like the figure below, using your IP address, username, and password.
For the Schema option, choose Updated, which creates a better TPC-C schema with more appropriate structure and better indexes.
- In this case, the Number of Warehouses (the scale) is set to 2000, but you don’t have to set it that high, because creating 2000 warehouses will take several hours to complete. Some guidelines suggest 10 to 100 warehouses per CPU. For this tutorial, set this value to 10 times the number of cores: 160 for a 16-core instance.
- For Virtual Users to Build Schema, choose a number that is between 1- and 2-times the number of client vCPUs. You can click the grey bar next to the slider to increment the number.
- Click OK
- Double click the Build option below the Schema Build section to create the schema and load the tables. When that completes, click the red flash light icon in the top center of the screen to destroy the virtual user and move to the next step.
If you created your database with the
Simple recovery model, you might
want to change it back to
Full at this point to get a more accurate test of a
production scenario. This will not take effect until after you take a full or
differential backup to trigger the start of the new log chain.
Creating the driver script
HammerDB uses the driver script to orchestrate the flow of SQL statements to the database to generate the required load.
- In the Benchmark panel, expand the Driver Script section and double-click Options.
- Verify the settings match what you used in the Schema Build dialog.
- Choose Timed Test Driver Script.
- The Checkpoint when complete option forces the database to write everything to disk at the end of the test, so check this only if you plan on running multiple tests in a row.
- To ensure a thorough test, set Minutes of Rampup Time to 5 and Minutes for Test Duration to 20.
- Click OK to exit the dialog.
- Double-click Load in the Driver Script section of the Benchmark panel to activate the driver script.
Creating virtual users
Creating a realistic load typically requires running scripts as multiple different users. Create some virtual users for the test.
- Expand the Virtual Users section and double click Options.
- If you set your warehouse count (scale) to 160, then set the Virtual Users to 16, because the TPC-C guidelines recommend a 10x ratio to prevent row locking. Select the Show Output checkbox to enable error messages in the console.
- Click OK
Collecting runtime statistics
HammerDB and SQL Server don’t easily collect detailed runtime statistics for
you. Although the statistics are available deep within SQL Server, they need to
be captured and calculated on a regular basis. If you do not already have a
procedure or tool to help capture this data, you can use the procedure below to
capture some useful metrics during your testing. The results will be written to
a CSV file in the Windows
temp directory. You can copy the data to a Google
Sheet using the Paste Special > Paste CSV option.
To use this procedure, you first must temporarily enable OLE Automation Procedures to write the file to disk,. Remember to disable it after testing:
sp_configure 'show advanced options', 1; GO RECONFIGURE; GO sp_configure 'Ole Automation Procedures', 1; GO RECONFIGURE; GO
Here’s the code to create the
sp_write_performance_counters procedure in SQL
Server Management Studio. Before starting the load test, you will execute this
procedure in Management Studio.:
USE [master] GO SET ANSI_NULLS ON GO SET QUOTED_IDENTIFIER ON GO /*** LogFile path has to be in a directory that SQL Server can Write To. */ CREATE PROCEDURE [dbo].[sp_write_performance_counters] @LogFile varchar (2000) = 'C:\\WINDOWS\\TEMP\\sqlPerf.log', @SecondsToRun int =1600, @RunIntervalSeconds int = 2 AS BEGIN --File writing variables DECLARE @OACreate INT, @OAFile INT, @FileName VARCHAR(2000), @RowText VARCHAR(500), @Loops int, @LoopCounter int, @WaitForSeconds varchar (10) --Variables to save last counter values DECLARE @LastTPS BIGINT, @LastLRS BIGINT, @LastLTS BIGINT, @LastLWS BIGINT, @LastNDS BIGINT, @LastAWT BIGINT, @LastAWT_Base BIGINT, @LastALWT BIGINT, @LastALWT_Base BIGINT --Variables to save current counter values DECLARE @TPS BIGINT, @Active BIGINT, @SCM BIGINT, @LRS BIGINT, @LTS BIGINT, @LWS BIGINT, @NDS BIGINT, @AWT BIGINT, @AWT_Base BIGINT, @ALWT BIGINT, @ALWT_Base BIGINT, @ALWT_DIV BIGINT, @AWT_DIV BIGINT SELECT @Loops = case when (@SecondsToRun % @RunIntervalSeconds) > 5 then @SecondsToRun / @RunIntervalSeconds + 1 else @SecondsToRun / @RunIntervalSeconds end SET @LoopCounter = 0 SELECT @WaitForSeconds = CONVERT(varchar, DATEADD(s, @RunIntervalSeconds , 0), 114) SELECT @FileName = @LogFile + FORMAT ( GETDATE(), '-MM-dd-yyyy_m', 'en-US' ) + '.txt' --Create the File Handler and Open the File EXECUTE sp_OACreate 'Scripting.FileSystemObject', @OACreate OUT EXECUTE sp_OAMethod @OACreate, 'OpenTextFile', @OAFile OUT, @FileName, 2, True, -2 --Write the Header EXECUTE sp_OAMethod @OAFile, 'WriteLine', NULL,'Transactions/sec, Active Transactions, SQL Cache Memory (KB), Lock Requests/sec, Lock Timeouts/sec, Lock Waits/sec, Number of Deadlocks/sec, Average Wait Time (ms), Average Latch Wait Time (ms)' --Collect Initial Sample Values SET ANSI_WARNINGS OFF SELECT @LastTPS= max(case when counter_name = 'Transactions/sec' then cntr_value end), @LastLRS = max(case when counter_name = 'Lock Requests/sec' then cntr_value end), @LastLTS = max(case when counter_name = 'Lock Timeouts/sec' then cntr_value end), @LastLWS = max(case when counter_name = 'Lock Waits/sec' then cntr_value end), @LastNDS = max(case when counter_name = 'Number of Deadlocks/sec' then cntr_value end), @LastAWT = max(case when counter_name = 'Average Wait Time (ms)' then cntr_value end), @LastAWT_Base = max(case when counter_name = 'Average Wait Time base' then cntr_value end), @LastALWT = max(case when counter_name = 'Average Latch Wait Time (ms)' then cntr_value end), @LastALWT_Base = max(case when counter_name = 'Average Latch Wait Time base' then cntr_value end) FROM sys.dm_os_performance_counters WHERE counter_name IN ( 'Transactions/sec', 'Lock Requests/sec', 'Lock Timeouts/sec', 'Lock Waits/sec', 'Number of Deadlocks/sec', 'Average Wait Time (ms)', 'Average Wait Time base', 'Average Latch Wait Time (ms)', 'Average Latch Wait Time base') AND instance_name IN( '_Total' ,'') SET ANSI_WARNINGS ON WHILE @LoopCounter <= @Loops BEGIN WAITFOR DELAY @WaitForSeconds SET ANSI_WARNINGS OFF SELECT @TPS= max(case when counter_name = 'Transactions/sec' then cntr_value end) , @Active = max(case when counter_name = 'Active Transactions' then cntr_value end) , @SCM = max(case when counter_name = 'SQL Cache Memory (KB)' then cntr_value end) , @LRS = max(case when counter_name = 'Lock Requests/sec' then cntr_value end) , @LTS = max(case when counter_name = 'Lock Timeouts/sec' then cntr_value end) , @LWS = max(case when counter_name = 'Lock Waits/sec' then cntr_value end) , @NDS = max(case when counter_name = 'Number of Deadlocks/sec' then cntr_value end) , @AWT = max(case when counter_name = 'Average Wait Time (ms)' then cntr_value end) , @AWT_Base = max(case when counter_name = 'Average Wait Time base' then cntr_value end) , @ALWT = max(case when counter_name = 'Average Latch Wait Time (ms)' then cntr_value end) , @ALWT_Base = max(case when counter_name = 'Average Latch Wait Time base' then cntr_value end) FROM sys.dm_os_performance_counters WHERE counter_name IN ( 'Transactions/sec', 'Active Transactions', 'SQL Cache Memory (KB)', 'Lock Requests/sec', 'Lock Timeouts/sec', 'Lock Waits/sec', 'Number of Deadlocks/sec', 'Average Wait Time (ms)', 'Average Wait Time base', 'Average Latch Wait Time (ms)', 'Average Latch Wait Time base') AND instance_name IN( '_Total' ,'') SET ANSI_WARNINGS ON SELECT @AWT_DIV = case when (@AWT_Base - @LastAWT_Base) > 0 then (@AWT_Base - @LastAWT_Base) else 1 end , @ALWT_DIV = case when (@ALWT_Base - @LastALWT_Base) > 0 then (@ALWT_Base - @LastALWT_Base) else 1 end SELECT @RowText = '' + convert(varchar, (@TPS - @LastTPS)/@RunIntervalSeconds) + ', ' + convert(varchar, @Active) + ', ' + convert(varchar, @SCM) + ', ' + convert(varchar, (@LRS - @LastLRS)/@RunIntervalSeconds) + ', ' + convert(varchar, (@LTS - @LastLTS)/@RunIntervalSeconds) + ', ' + convert(varchar, (@LWS - @LastLWS)/@RunIntervalSeconds) + ', ' + convert(varchar, (@NDS - @LastNDS)/@RunIntervalSeconds) + ', ' + convert(varchar, (@AWT - @LastAWT)/@AWT_DIV) + ', ' + convert(varchar, (@ALWT - @LastALWT)/@ALWT_DIV) SELECT @LastTPS = @TPS, @LastLRS = @LRS, @LastLTS = @LTS, @LastLWS = @LWS, @LastNDS = @NDS, @LastAWT = @AWT, @LastAWT_Base = @AWT_Base, @LastALWT = @ALWT, @LastALWT_Base = @ALWT_Base EXECUTE sp_OAMethod @OAFile, 'WriteLine', Null, @RowText SET @LoopCounter = @LoopCounter + 1 END --CLEAN UP EXECUTE sp_OADestroy @OAFile EXECUTE sp_OADestroy @OACreate print 'Completed Logging Performance Metrics to file: ' + @FileName END GO
Running the TPC-C load test
In SQL Server Management Studio, execute the collection procedure using the following script:
Use master Go exec dbo.sp_write_performance_counters
On the Compute Engine instance where you installed HammerDB, start the test in the HammerDB application:
- In the Benchmark panel, under Virtual Users double-click Create to create the virtual users, which will activate the Virtual User Output tab.
- Double-click Run just below the Create option to kick off the test.
- When the test completes you will see the Transactions Per Minute (TPM) calculation in the Virtual User Output tab.
- You can find the results from your collection procedure in the
- Save all of these values to a Google Sheet and use them to compare multiple test runs.
After you've finished the SQL Server load-testing tutorial, you can clean up the resources you created on Google Cloud Platform so you won't be billed for them in the future. The following sections describe how to delete or turn off these resources.
Deleting the project
The easiest way to eliminate billing is to delete the project you created for the tutorial.
To delete the project:
- In the GCP Console, go to the Projects page.
- In the project list, select the project you want to delete and click Delete project.
- In the dialog, type the project ID, and then click Shut down to delete the project.
To delete a Compute Engine instance:
- In the GCP Console, go to the VM Instances page.
- Click the checkbox next to the instance you want to delete.
- Click the Delete button at the top of the page to delete the instance.