SQL Server 2016 tempdb configurations benchmark (SQL Server Series)

26 June 2016 Hans Michiels 10 Comments

Introduction

When you have installed SQL Server 2016 you will have noticed that the configuration for tempdb has changed.
Microsoft made a proper tempdb configuration easy to do via the GUI during installation.

Today I want to go through the settings offered and will measure performance of different setups.

What are the options?

During installation you can set the following values:
For TempDB datafiles:

Setting	Default value
Number of files	4
Initial size (MB)	8
Autogrowth (MB)	64
Data directories	Default data directory, in my case C:\MSSQL\Instance root\MSSQL13.MSSQLSERVER\MSSQL\Data

For TempDB logfile:

Setting	Default value
Initial size (MB)	8
Autogrowth (MB)	64
Log directory	Default data directory, in my case C:\MSSQL\Instance root\MSSQL13.MSSQLSERVER\MSSQL\Data

MSSQL-S01E03-tempdb-setup
Configuration screen during SQL Server 2016 installation (I changed the directories already).

So what is the best setup?

This subject is not new for SQL Server 2016, but the installation procedure now just gives a better opportunity to set it up right the first time.
There are roughly five differentiators in the setup:

Initial size (MB)
Autogrowth (MB)
Number of data files
Location of data files
Read/write performance of the different storage media

What are the considerations when setting these values?

Initial size (MB)

A larger initial size will accomplish two things:

It will take longer before the file will have the need to grow.
The file can be stored less fragmented on disk.

If you have enough disk space, the larger the better, there is no real downside on making the initial size quite large. More about this in my conclusion.

Autogrowth (MB)

With autogrowth the choice is a bit more delicated, as a small or large value will both have pros and cons, as pointed out in the schedule below.

Autogrowth value	Pros	Cons
Small	More frequent smaller growth actions will cause less noticable hick ups	TempDB datafiles will become more fragmented.
Large	Tempdb datafiles will become less fragmented.	Less frequent larger growth actions might cause noticable hick ups

Number of data files

In general, the recommendation is one data file per (logical) processor. You can configure more files than the number of (logical) processors, but do this only if you notice that this gives a better performance.

Location of data files

Locating the data files on different drives can have two effects:

Queue length per disk drive in Windows (all scenarios, also with virtual disks residing on the same physical drive).
Physical writes to the diskdrive can go in parallel (only when different physical drives).

The tests

The testcases

Creating combinations of all differentiators would just create *a lot* of testscenarios.
So I will be a bit pragmatic.
The testcases are chosen, so that they can be compared with at least one other testcase, of which one differentiator is different.

Test #	# Data files	Init size (MB)	Auto growth (MB)	# Fol- ders	Scenario
0	4	8 (x4)	64	1	Default installation (data files on OS disk)
1	4	8 (x4)	64	1	Data files on dedicated drive
2	4	8 (x4)	64	4	Data files on more drives
3	4	2048 (x4)	64	4	Data files on more drives + Larger initial size
4	4	8 (x4)	256	4	Data files on more drives + Larger autogrowth
5	4	2048 (x4)	64	1	Data files on dedicated drive + Larger initial size
6	4	2048 (x4)	64	1	Data files on dedicated drive + Larger initial size + TempDB on different SSD
7	8	2048 (x8)	64	1	Data files on dedicated drive + Larger initial size + More data files
8	4	8 (x4)	64	1	Default installation (data files on OS disk) + Larger initial size

Test results

I will present you both two graphs as well as the raw data. I think especially the first graph gives some useful insights, as it compares all testcases with testcase 0 (default installation), by using an index 100 for testcase 0. If a different testcase performs better, the index will be less than 100. Also, in this way, the average result of both testsets (A and B) can be shown.
I left testcase 6 out of the graph, later I will tell why I did this.
MSSQL-S01E03-testresult-index Index-based comparison of testresults

MSSQL-S01E03-testresult-seconds Execution times of testcases for testset A and B

Test results figures

Test #	Scenario	Testresult testset A (ms)	Testresult testset B (ms)
0	Default installation	30359	232795
1	Data files on dedicated drive	28344	261467
2	Data files on more drives	22514	248484
3	Data files on more drives + Larger initial size	28040	210223
4	Data files on more drives + Larger autogrowth	24173	200632
5	Data files on dedicated drive + Larger initial size	19608	192587
6	Data files on dedicated drive + Larger initial size + TempDB on different SSD	123289	387646
7	Data files on dedicated drive + Larger initial size + More data files	19692	211501
8	Default installation + Larger initial size	20149	361844

Conclusion

Data files on dedicated drive + Larger initial size (test 5) gives the best performance for both testsets. Putting tempdb files on multiple disks can not win from this, however the differences are relatively small.

Surprisingly performance dramaticly decreases when the virtual disk on which tempdb is stored is moved to a different (faster) SSD (this is test 6)! I have no explanation for this. Therefore I will leave test 6 out of the analysis, by doing this physical storage is the same for all testcases and changes in performance are only caused by the other differentiators.

My most important conclusion is that the default values offered during installation are quite okay, except for two things:

The Initial size (MB)

You should really set the initial size to a big number (gigabytes if possible). Note that if you leave the number of files 4 (which I recommend), the space used is four times the size that you use for Initial size.
If you have a dedicated disk for tempdb (which is recommended), you might want to set the initial size of the datafile(s) to fill up the disk. You should then disable Autogrowth.
Also on a production server, check the size of tempdb files after a week of uptime (also the SQL instance may not have been restarted). Say the tempdb size of all files together (can also be one file) is 22GB. You could then set your initial size to 28GB in total, so 7GB per file if you configured 4 files. This would mean that during normal operation, the tempdb never has to grow.

Dedicated disk

It is a common best practive to put tempdb on a seperate disk so that the Windows Disk Queue does not have to compete with other read/write actions to that disk.

Other settings are somewhat less important but … think for yourself (my disclaimer)

In my opinion both putting tempdb files on multiple different disks and setting Autogrowth to a different value do not offer much performance benefit.
The Autogrowth value is even irrelevant if the initial size is so large that temdb never has to grow during normal operation.
But every server is different! Test and think for yourself.
My testresults are presented “AS IS” in the hope they will be useful as a starting point for further own analysis. I am not responsible for any disadvantage you might have by using my test results for your tempdb setup.

What if SQL Server is already installed and/or it is an older version?

No worries, you still can change the tempdb configuration.
You can move and add files using SQL scripts.
Initial size and Autogrowth you can also set with Management Studio.
MSSQL-S01E03-166 Go to properties of tempdb

MSSQL-S01E03-167 Here you can change Initial size and Autogrowth

By the way, I noticed that SQL Server does not delete moved tempdb files. So if you move tempdb files, you should delete the files on the old location, so that they do not use disk space and do not cause confusion for administrators.

Test justification

All tests are done on a Hyper V Virtual Machine running Windows Server 2012 R2 and SQL Server 2016 Developer Edition.
Change of physical storage type is done by moving vhdx files to different physical disks, and then change the file location in the settings of the Hyper V Virtual Machine (while it is not running, of course).
Each test is done on an initial copy of the Virtual Machine, therewith the startsituation is equal for every test (e.g. no tempdb growth by previous tests).
Tests where done with two sets of scripts (set A and B) that inserted, updated and deleted rows in temporary tables in tempdb. Set A consists of three sql scripts, and set B of five.
All scripts were developed, so that all tempdb inserts started at the same time (maybe milliseconds difference in starttime only). This to ensure that every testrun was executed in the same way.
All scripts were executed in SQL Server Management Studio.
The first file of tempdb is stored on its own (virtual) disk except for testcase 0, where tempdb files are on the C-drive where Windows is installed.
When spreading files over multiple virtual disks, the disks created for database files, log files, and backup files are used. This was done for pragmatic reasons during the test. This was defendable, because no other activity on databases and log files was going on during the test. It is however not recommended to spread tempdb like that in production environment, unless every disk is dedicated to be used for tempdb only.
The host machine had the same applications open during every test, being a Windows Explorer and Microsoft Excel. No other applications that used CPU and could influence performance of the virtual machines were running.
The tempdb log file was placed on a seperate disk (the same one) for each test. Therewith it was no subject of performance investigation. This would have at least doubled the number of testcases.

Wrap up

Microsoft did a good job in improving the setup options for tempdb in the installation procedure.
It is a bit of a pity that they did not go the last mile, giving the tempdb data files a more realistic Initial size.
Luckily this is easy to change during installation, and it is a quick win for anyone installing SQL Server 2016.

Download all the used scripts here.

SSIS Series

Zip sourcefiles in a SSIS Package (SSIS Series)

17 June 2016 Hans Michiels

Introduction

As I am in a (un)zipping mood, this week the opposite problem of last week.
The example is not so ready-to-use as in my previous post, as it depends on choices made in your development process of other SSIS packages.
Still I hope you can adjust this example to match your needs.

Problem

If you have imported and archived (csv- or xml) files, these files can take up a lot of disk space. The archive folder can also become overwhelmingly full with ten thousands of files, which can make it slowly accessible and difficult to manage.

Solution

To tackle this problem I have made a fully configurable SSIS package, that can be executed from a different package with an Execute Package Task.
The package is made using SSIS 2014.

The zipping itself is done with 7za.exe, for more information on how to get and install this free program see my previous blog post.

Scenario

When I used this concept, a different SSIS package moved files to a subfolder of an archive folder after they were processed, but their name was prefixed with the LoadCycleID with leading zeros, for instance 001345_customers.xml.

The archive folder and subfolder were separated, because the archive folder was an environment variable and the same for all packages, while the subfolder was different for different SSIS packages, and contained the name of the source system of the files.
If you only have one archive folder, just change the example to use that instead of the concatenation of archive folder and subfolder.

Overview

Here is a picture of the end result:

SSIS-S01E06-124
The Control Flow

Detailed steps

Parameters and variables

First you need to add a few parameters and variables to the package.

Parameters:

Name	Data type	Required	Description
_7zaPathAndFilename	String	False	The full path and filename of 7za.exe.
ArchiveFolder	String	True	The root folder for archiving processed files.
ArchiveSubFolder	String	True	The subfolder of the ArchiveFolder in which the files to be zipped exist, the subfolder can for instance be named after a source system like SAP or CRM.
LoadCycleID	Integer	True	The ID of the LoadCycle that is processing these files. In your situation it could also be called ExecutionID, SessionID and so on, the name is not really important.

SSIS-S01E06-125
Package parameters

Variables:

However not all variables might be necessary, I like the concept of being able to configure all values in one place.
The data type of all variables is String, except LoadCycleID which is an Integer.
Every variable uses an expression to set its value.
Here is a detailed list of all variables and the expression used for each one.

Name	Expression
ArchiveFilesToDeleteInFolder	@[$Package::ArchiveFolder] + "\\" + @[$Package::ArchiveSubFolder]
ArchiveFilesToDeletePattern	@[User::LoadCycleIDString] + "*"
ArchiveFilesToZipPattern	@[$Package::ArchiveFolder] + "\\" + @[$Package::ArchiveSubFolder] + "\\" + @[User::LoadCycleIDString] + "."
ArchiveFilesWorkingDirectory	@[$Package::ArchiveFolder]
ArchiveZipFile	@[User::LoadCycleIDString] + "_" + @[$Package::ArchiveSubFolder] + ".zip"
ArchiveZipPathFile_Final	@[$Package::ArchiveFolder] + "\\" +@[$Package::ArchiveSubFolder] + "\\" + @[User::LoadCycleIDString] + "_" + @[$Package::ArchiveSubFolder] + ".zip"
ArchiveZipPathFile_Temp	@[$Package::ArchiveFolder] + "\\" + @[User::LoadCycleIDString] + "_" + @[$Package::ArchiveSubFolder] + ".zip"
LoadCycleID	@[$Package::LoadCycleID]
LoadCycleIDString	RIGHT("00000" + (DT_STR, 10, 1252) @[$Package::LoadCycleID], 6)

Important note: if you expect that the value for LoadCycleID will grow beyond 999999 over time, you should add more prefixed zero’s to LoadCycleIDString and increase the number 6, the second parameter of the RIGHT function.

SSIS-S01E06-126
SSIS variables of the package

The Control Flow

The control flow is quite straight forward, only three tasks and we’re done.

EPR Zip files in archive subfolder

This task creates one zipfile containing multiple other files in the ArchiveSubFolder.
From the SSIS Toolbox drag an Execute Process Task to the Control Flow panel and open the Execute Process Task Editor.
On the General and Process panes you do not have to change anything.

SSIS-S01E06-128
SSIS-S01E06-129
No changes needed here

On the Expressions pane, press the ellipsis next to Expressions. In the Property Expressions Editor add the following:

Property	Expression
Arguments	"a "+ "\"" + @[User::ArchiveZipFile] + "\" " + "\"" + @[User::ArchiveFilesToZipPattern] + "\" -mx=9"
Executable	@[$Package::_7zaPathAndFilename]
WorkingDirectory	@[User::ArchiveFilesWorkingDirectory]

Please note that in the Arguments expression -mx=9 is used. This is the compression level, level 9 means “ultra” compression. This level gives the smallest zip files but is also the slowest compression level. You can for instance also use 5 for “normal” or 3 for “fast” compression, if this is what you want.

SSIS-S01E06-127
Expressions for the Execute Process Task

SCR Delete uncompressed files

This task deletes the files that were added to the zip file in the previous task.

In the Script Task Editor, add both variables User::ArchiveFilesToDeleteInFolder and User::ArchiveFilesToDeletePattern to ReadOnlyVariables, then press Edit Script ….

Then in the namespaces region add one namespace, as indicated below.

Namespaces region

#region Namespaces
using System;
using System.Data;
using Microsoft.SqlServer.Dts.Runtime;
using System.Windows.Forms;
// Added namespace:
using System.IO;
#endregion

Adjust public void main() as follows:

public void Main

public void Main()
{
  // Copy SSIS variable values to C# string variables.
  string archiveFilesToDeleteInFolder = Dts.Variables["User::ArchiveFilesToDeleteInFolder"].Value.ToString();
  string archiveFilesToDeletePattern = Dts.Variables["User::ArchiveFilesToDeletePattern"].Value.ToString();

  // Delete each file that just has been added to the zip archive.
  foreach (string f in Directory.EnumerateFiles(archiveFilesToDeleteInFolder, archiveFilesToDeletePattern))
  {
    File.Delete(f);
  }

Dts.TaskResult = (int)ScriptResults.Success;
}

WARNING: Make sure that no files that meet the file name pattern of files to add to the zip file are added to the ArchiveSubFolder while or after the zipfile is (being) created. Otherwise files that are NOT zipped might get deleted!

Download the script here.

FST Move zip file

The File System Task is to move the zip file to the ArchiveSubFolder, because the zipfile is initially created in a temporary folder and not in the ArchiveSubFolder.
This is to prevent that it would be deleted by coincidence, if you use a value for ArchiveFilesToDeletePattern that ends with .*
Configure the File System Task as shown in the picture below:

SSIS-S01E06-130
File System Task Configuration

Download the script here.

Let the car drive

Time to test the package and tell the result in pictures (if you don’t mind).

SSIS-S01E06-131
Before: files of LoadCycleID 001346 need to be zipped ..

SSIS-S01E06-132
Execute the package ..

SSIS-S01E06-133
All csv files replaced with one zip file ..

SSIS-S01E06-134
.. that contains the csv files ..

Conclusion / Wrap up

In this blog post you could read the steps needed to build a SSIS package that creates a zip file containing a number of files in an archive(sub)folder.
This makes the archive smaller and easier to manage.
However beware that you only delete files that were added to the zip file. For more details see the warning above.

SSIS Series

Unzip specific files from lots of zipped archives (SSIS Series)

12 June 2016 Hans Michiels 1 Comment

Introduction

This week I want to discuss the trouble you may have when searching hundreds of zip files for specific files inside them.
Recognizable?
At least I faced this problem some time ago, and solved it using SSIS and the commandline version of 7-zip, named 7za.exe.
The choice for SSIS was purely based on convenience, as you could tackle this problem also using Powershell, C# or almost any other scripting language.
But I thought SSIS is just a neat and quick way to do this.

Problem

Finding specific files in huge amounts of zip files.

Solution

To tackle this problem I have made a fully configurable SSIS package, that can be executed with different parameters.
The main task of the package is to extract specific files (based on a file name pattern) from zipfiles in a folder and – optionally – subfolders.
The unzipping itself is done with 7za.exe, which is the commandline version of 7z, the great open source zip program, that exists already for several years and is used by millions of users.

By the way, the package is made using SSIS 2014.

Preparations: download and install 7za

It is important to download this package from the official 7z web site, just to avoid that you would get a copy from some deceitful web site infected with malware.
To go to the official 7-zip download page click on this link:
7-zip download page
Most important is the 7-Zip Extra: standalone console version.
For the example I have used version 16.02 (2016-05-21) for Windows, but it should also work with version 9.20 (2010-11-18), then called 7-Zip Command Line Version.

If you download the newest version, you also need software to unzip the 7z160-extra.7z file, for instance the regular 32-bit or 64-bit version of 7-zip, which can also be downloaded on the same page.

SSIS-S01E05-100
Download page of 7-zip

I assume you will use the newest version 7z160-extra.7z
Once unzipped, just take the time to read License.txt, readme.txt and history.txt.
Then copy the three files 7zxa.dll, 7za.dll and 7za.exe to a folder on the Windows computer where SSIS runs, usually a Windows Server.
If you run a x64 version of Windows, you can take the x64 version of the 7za files, which are in a subfolder “x64” of the unzipped files.

SSIS-S01E05-101
Unzipped 7za files ..

I have created a subfolder “7za” in the “C:\Program files” folder of my virtual Windows Server, and copied the x64-version of the three mentioned files to it, but you can put them in any folder you like.
When you have done this, the real fun can start ..