Difference between revisions of "Image Compression Tests"

From FOG Project
Jump to: navigation, search
m (Added small intro at top.)
(Wayne Workman)
 
(4 intermediate revisions by the same user not shown)
Line 5: Line 5:
 
== devinR ==
 
== devinR ==
  
This isn’t strictly a tutorial, but I am sharing info, so here we go! I loved the work by Ch3i here: https://forums.fogproject.org/topic/4948/compression-tests, but I wanted to know a bit more. Yesterday I spent more time than I should tell my boss just uploading the same image with different compression ratings. Here’s a brief table of contents:
+
This isn’t strictly a tutorial, but I am sharing info, so here we go! I loved the work by Ch3i here: https://forums.fogproject.org/topic/4948/compression-tests, but I wanted to know a bit more. Yesterday I spent more time than I should tell my boss just capturing the same image with different compression ratings. Here’s a brief table of contents:
  
 
'''Questions'''
 
'''Questions'''
Line 20: Line 20:
  
 
*How much extra space does each compression level take (compared to max 9)?
 
*How much extra space does each compression level take (compared to max 9)?
*How long does each compression level take to upload?
+
*How long does each compression level take to capture?
*How long does each compression level take to download?
+
*How long does each compression level take to deploy?
 
*What is best for my environment?
 
*What is best for my environment?
  
Line 35: Line 35:
 
FOG server (1.2.0) is a dual core 2.8GHz dell Optiplex 380 (2008ish) with Debian 7 (Wheezy). It has 1x7200RPM HDD (80GB) as the linux root file system with a pair of WD “Black” 1TB 7200RPM HDDs in a linux software raid mirror mounted on /images. It also acts as our DHCP server for known computers, unknown devices like phones or tablets, and VoIP phones (between 350 and 400 devices during the business day). Connected to netgear unmanaged Gig switch with a single link.
 
FOG server (1.2.0) is a dual core 2.8GHz dell Optiplex 380 (2008ish) with Debian 7 (Wheezy). It has 1x7200RPM HDD (80GB) as the linux root file system with a pair of WD “Black” 1TB 7200RPM HDDs in a linux software raid mirror mounted on /images. It also acts as our DHCP server for known computers, unknown devices like phones or tablets, and VoIP phones (between 350 and 400 devices during the business day). Connected to netgear unmanaged Gig switch with a single link.
  
FOG client is a new dell Optiplex 3020 with the dell image (with the first 2 partitions removed. Only a single partition (sda3) is being uploaded\downloaded. This client has something weird with for server, so all uploads and downloads cause the client to take like 80s with a blinking cursor before it fully loads and continues. This makes all of my times show longer and all of my speeds to be lower than actually observed. Connected to same netgear unmanaged Gig switch with a single link.
+
FOG client is a new dell Optiplex 3020 with the dell image (with the first 2 partitions removed. Only a single partition (sda3) is being captured\deployed. This client has something weird with for server, so all captures and deploys cause the client to take like 80s with a blinking cursor before it fully loads and continues. This makes all of my times show longer and all of my speeds to be lower than actually observed. Connected to same netgear unmanaged Gig switch with a single link.
  
All images were sent with unicast download tasks and upload were done with a Single disk multi partition fixed size option set.
+
All images were sent with unicast deploy tasks and capture were done with a Single disk multi partition fixed size option set.
  
A few times I had to run the upload twice because I didn’t save the image the first time I uploaded. This is why there are two entries for uploads. All uploads were completed before any downloads to make sure that we didn’t get a chain effect of errors from 1 compression to another.
+
A few times I had to run the capture twice because I didn’t save the image the first time I captured. This is why there are two entries for captures. All captures were completed before any deployments to make sure that we didn’t get a chain effect of errors from 1 compression to another.
  
The raid utilization was determined by polling the raid hard drives active time every second for a duration of 100 seconds during the download. This was accomplished with the command “iostat -dx /dev/sdb /dev/sdc 1 100 > Compression5.log” I then took these 100 entries and averaged them to produce the snapshot of about 1/4th of the total download time for the image.
+
The raid utilization was determined by polling the raid hard drives active time every second for a duration of 100 seconds during the deploy. This was accomplished with the command “iostat -dx /dev/sdb /dev/sdc 1 100 > Compression5.log” I then took these 100 entries and averaged them to produce the snapshot of about 1/4th of the total deployment time for the image.
  
 
'''Conclusions'''
 
'''Conclusions'''
  
*There are bigger factors for download time than compression. I was concerned about why compress 5 had such bad download performance. I ran the test again and got much better results. It just seems that the FOG server was busier, or our network was under heavy load, causing many packet failures, or something weird. I don’t know what. This makes believe that while greener is better for download duration, at any time, any of the higher compression images can have worse time to download times than a lower compression.
+
*There are bigger factors for deployment time than compression. I was concerned about why compress 5 had such bad deploy performance. I ran the test again and got much better results. It just seems that the FOG server was busier, or our network was under heavy load, causing many packet failures, or something weird. I don’t know what. This makes believe that while greener is better for deployment duration, at any time, any of the higher compression images can have worse time to deployment times than a lower compression.
  
*Upload time can safely be reduced with minimal impact on image size. Even if it is just to compression 8, time goes to 75% of compression 9. Going to compression 7 further lowers it to about 60% of the time with a .2% increase in stored image size.
+
*Capture time can safely be reduced with minimal impact on image size. Even if it is just to compression 8, time goes to 75% of compression 9. Going to compression 7 further lowers it to about 60% of the time with a .2% increase in stored image size.
  
*I care about deployment time, so any compression level above 1 will give me a good deployment time. I also like my upload to not take forever, so going forward, my compression level is getting set to 7.
+
*I care about deployment time, so any compression level above 1 will give me a good deployment time. I also like my capture to not take forever, so going forward, my compression level is getting set to 7.
  
  
 
'''Future work'''
 
'''Future work'''
  
*Same download tests but multicasting
+
*Same deployment tests but multicasting
*Same download tests but to multiple hosts at once (4 or 5)
+
*Same deployment tests but to multiple hosts at once (4 or 5)
 
*Other?
 
*Other?
  
Line 90: Line 90:
 
Tasks :
 
Tasks :
  
Upload : not tested
+
Capture : not tested
 
Donwload : ~3G/Min - ~29 mins
 
Donwload : ~3G/Min - ~29 mins
 
Compression : 5 ( '''The winner''' :p )
 
Compression : 5 ( '''The winner''' :p )
Line 102: Line 102:
 
Tasks :
 
Tasks :
  
Upload : ~4.9G/Min - ~17 mins
+
Capture : ~4.9G/Min - ~17 mins
 
Donwload : ~6.4G/Min - ~13mins
 
Donwload : ~6.4G/Min - ~13mins
 
Compression : 9
 
Compression : 9
Line 114: Line 114:
 
Tasks :
 
Tasks :
  
Upload : ~1.9G/Min - ~43mins
+
Capture : ~1.9G/Min - ~43mins
 
Donwload : ~6.9G/Min - ~12mins
 
Donwload : ~6.9G/Min - ~12mins
 +
 +
== Wayne Workman ==
 +
 +
The below tests were performed with the FOG working branch just after the 1.4.1 release, commit <font color="red">b4544755284c6914fb880e4d41b45c3a6ca41d57</font>. The image being used was a Windows 10 image, approximately 11GB in size uncompressed. All the testhosts have 3 cores, 4GB of RAM, and SSD disks. The fog server has 1 core, 1GB of RAM, and a mechanical disk. The image was always set to “Single Disk - Resizable.” There are no hops between the clients and the fog server, and there were no other loads on the fog server or clients or other physical hardware. All test results were achieved programmatically.
 +
 +
'''Partimage - compression 3'''
 +
 +
Image capture of “testHost1” completed in about “4” minutes.
 +
 +
Completed image deployment to “testHost1” in about “3” minutes.
 +
 +
Completed image deployment to “testHost2” in about “5” minutes.
 +
 +
Completed image deployment to “testHost3” in about “7” minutes.
 +
 +
All image deployments completed in about “9” minutes.
 +
 +
 +
'''Partclone gzip - compression 3'''
 +
 +
Image capture of “testHost1” completed in about “4” minutes.
 +
 +
Completed image deployment to “testHost1” in about “3” minutes.
 +
 +
Completed image deployment to “testHost2” in about “5” minutes.
 +
 +
Completed image deployment to “testHost3” in about “7” minutes.
 +
 +
All image deployments completed in about “9” minutes.
 +
 +
 +
'''Partclone zstd - compression 3'''
 +
 +
Image capture of “testHost1” completed in about “3” minutes.
 +
 +
Completed image deployment to “testHost1” in about “3” minutes.
 +
 +
Completed image deployment to “testHost2” in about “5” minutes.
 +
 +
Completed image deployment to “testHost3” in about “7” minutes.
 +
 +
All image deployments completed in about “9” minutes.
 +
 +
 +
Just to point out - it’s not fair to compare zstd with gzip or pigz. This is because zstd’s compression scale goes from 1 to 22 while gzip & pigz compression scale is from 1 to 9 I think. Having gzip and pigz at 3 is much higher compression than zstd at 3. In effect, pigz and gzip were at about 30% compression while zstd was at about 15% compression. Yet zstd still outperformed. So for fun, I ran the below tests.
 +
 +
 +
'''Partclone zstd - compression 12'''
 +
 +
Image capture of “testHost1” completed in about “5” minutes.
 +
 +
Completed image deployment to “testHost1” in about “3” minutes.
 +
 +
Completed image deployment to “testHost2” in about “4” minutes.
 +
 +
Completed image deployment to “testHost3” in about “6” minutes.
 +
 +
All image deployments completed in about “8” minutes.
 +
 +
 +
'''Partclone zstd - compression 17'''
 +
 +
Image capture of “testHost1” completed in about “13” minutes.
 +
 +
Completed image deployment to “testHost1” in about “3” minutes.
 +
 +
Completed image deployment to “testHost2” in about “4” minutes.
 +
 +
Completed image deployment to “testHost3” in about “6” minutes.
 +
 +
All image deployments completed in about “8” minutes.
 +
 +
 +
Notes on my findings - an 11GB image is small for a Windows image. Typically Windows images for actual production deployment are larger, generally from 20GB to 30GB (in 2016). I’ve seen people deploy images as large as 160GB though. The larger the images are, the more ZSTD will pull ahead of the others and have a clear-cut superior performance. All research on it shows this, and big images in FOG will show it, too.

Latest revision as of 03:56, 2 June 2017

Below are some studies and conclusions conducted by the FOG community's users.


devinR

This isn’t strictly a tutorial, but I am sharing info, so here we go! I loved the work by Ch3i here: https://forums.fogproject.org/topic/4948/compression-tests, but I wanted to know a bit more. Yesterday I spent more time than I should tell my boss just capturing the same image with different compression ratings. Here’s a brief table of contents:

Questions

  • Data
  • Methodology\Specs
  • Conclusions
  • Future work

So without further delay, lets go!

Questions

Here are some questions that I plan to answer:

  • How much extra space does each compression level take (compared to max 9)?
  • How long does each compression level take to capture?
  • How long does each compression level take to deploy?
  • What is best for my environment?

Data

Fog-compression.png

Picture of my data, yay!


Methodology\Specs

FOG server (1.2.0) is a dual core 2.8GHz dell Optiplex 380 (2008ish) with Debian 7 (Wheezy). It has 1x7200RPM HDD (80GB) as the linux root file system with a pair of WD “Black” 1TB 7200RPM HDDs in a linux software raid mirror mounted on /images. It also acts as our DHCP server for known computers, unknown devices like phones or tablets, and VoIP phones (between 350 and 400 devices during the business day). Connected to netgear unmanaged Gig switch with a single link.

FOG client is a new dell Optiplex 3020 with the dell image (with the first 2 partitions removed. Only a single partition (sda3) is being captured\deployed. This client has something weird with for server, so all captures and deploys cause the client to take like 80s with a blinking cursor before it fully loads and continues. This makes all of my times show longer and all of my speeds to be lower than actually observed. Connected to same netgear unmanaged Gig switch with a single link.

All images were sent with unicast deploy tasks and capture were done with a Single disk multi partition fixed size option set.

A few times I had to run the capture twice because I didn’t save the image the first time I captured. This is why there are two entries for captures. All captures were completed before any deployments to make sure that we didn’t get a chain effect of errors from 1 compression to another.

The raid utilization was determined by polling the raid hard drives active time every second for a duration of 100 seconds during the deploy. This was accomplished with the command “iostat -dx /dev/sdb /dev/sdc 1 100 > Compression5.log” I then took these 100 entries and averaged them to produce the snapshot of about 1/4th of the total deployment time for the image.

Conclusions

  • There are bigger factors for deployment time than compression. I was concerned about why compress 5 had such bad deploy performance. I ran the test again and got much better results. It just seems that the FOG server was busier, or our network was under heavy load, causing many packet failures, or something weird. I don’t know what. This makes believe that while greener is better for deployment duration, at any time, any of the higher compression images can have worse time to deployment times than a lower compression.
  • Capture time can safely be reduced with minimal impact on image size. Even if it is just to compression 8, time goes to 75% of compression 9. Going to compression 7 further lowers it to about 60% of the time with a .2% increase in stored image size.
  • I care about deployment time, so any compression level above 1 will give me a good deployment time. I also like my capture to not take forever, so going forward, my compression level is getting set to 7.


Future work

  • Same deployment tests but multicasting
  • Same deployment tests but to multiple hosts at once (4 or 5)
  • Other?


ch3i

I have try some compression level, below results :

Host configuration :

Core I5 12 Go RAM Ethernet : Giga LAN Configuration :

Ethernet : Giga Deploy method : unicast

Installation :

Single boot Windows 7 Size : 87 Go Compression : 0

-rwxrwxrwx 1 root root 512 févr. 10 10:32 d1.mbr -rwxrwxrwx 1 root root 0 févr. 10 10:32 d1.original.swapuuids -rwxrwxrwx 1 root root 25363450 févr. 10 10:32 d1p1.img -rwxrwxrwx 1 root root 86580172755 févr. 10 10:53 d1p2.img -rwxrwxrwx 1 root root 171142723 févr. 10 10:54 d1p3.img

Tasks :

Capture : not tested Donwload : ~3G/Min - ~29 mins Compression : 5 ( The winner :p )

-rwxrwxrwx 1 root root 512 mai 13 10:45 d1.mbr -rwxrwxrwx 1 root root 0 mai 13 10:45 d1.original.swapuuids -rwxrwxrwx 1 root root 8544045 mai 13 10:45 d1p1.img -rwxrwxrwx 1 root root 40054036052 mai 13 11:03 d1p2.img -rwxrwxrwx 1 root root 3459293 mai 13 11:04 d1p3.img

Tasks :

Capture : ~4.9G/Min - ~17 mins Donwload : ~6.4G/Min - ~13mins Compression : 9

-rwxrwxrwx 1 root root 512 mai 13 11:26 d1.mbr -rwxrwxrwx 1 root root 0 mai 13 11:26 d1.original.swapuuids -rwxrwxrwx 1 root root 8471371 mai 13 11:26 d1p1.img -rwxrwxrwx 1 root root 39740659244 mai 13 12:10 d1p2.img -rwxrwxrwx 1 root root 3035522 mai 13 12:10 d1p3.img

Tasks :

Capture : ~1.9G/Min - ~43mins Donwload : ~6.9G/Min - ~12mins

Wayne Workman

The below tests were performed with the FOG working branch just after the 1.4.1 release, commit b4544755284c6914fb880e4d41b45c3a6ca41d57. The image being used was a Windows 10 image, approximately 11GB in size uncompressed. All the testhosts have 3 cores, 4GB of RAM, and SSD disks. The fog server has 1 core, 1GB of RAM, and a mechanical disk. The image was always set to “Single Disk - Resizable.” There are no hops between the clients and the fog server, and there were no other loads on the fog server or clients or other physical hardware. All test results were achieved programmatically.

Partimage - compression 3

Image capture of “testHost1” completed in about “4” minutes.

Completed image deployment to “testHost1” in about “3” minutes.

Completed image deployment to “testHost2” in about “5” minutes.

Completed image deployment to “testHost3” in about “7” minutes.

All image deployments completed in about “9” minutes.


Partclone gzip - compression 3

Image capture of “testHost1” completed in about “4” minutes.

Completed image deployment to “testHost1” in about “3” minutes.

Completed image deployment to “testHost2” in about “5” minutes.

Completed image deployment to “testHost3” in about “7” minutes.

All image deployments completed in about “9” minutes.


Partclone zstd - compression 3

Image capture of “testHost1” completed in about “3” minutes.

Completed image deployment to “testHost1” in about “3” minutes.

Completed image deployment to “testHost2” in about “5” minutes.

Completed image deployment to “testHost3” in about “7” minutes.

All image deployments completed in about “9” minutes.


Just to point out - it’s not fair to compare zstd with gzip or pigz. This is because zstd’s compression scale goes from 1 to 22 while gzip & pigz compression scale is from 1 to 9 I think. Having gzip and pigz at 3 is much higher compression than zstd at 3. In effect, pigz and gzip were at about 30% compression while zstd was at about 15% compression. Yet zstd still outperformed. So for fun, I ran the below tests.


Partclone zstd - compression 12

Image capture of “testHost1” completed in about “5” minutes.

Completed image deployment to “testHost1” in about “3” minutes.

Completed image deployment to “testHost2” in about “4” minutes.

Completed image deployment to “testHost3” in about “6” minutes.

All image deployments completed in about “8” minutes.


Partclone zstd - compression 17

Image capture of “testHost1” completed in about “13” minutes.

Completed image deployment to “testHost1” in about “3” minutes.

Completed image deployment to “testHost2” in about “4” minutes.

Completed image deployment to “testHost3” in about “6” minutes.

All image deployments completed in about “8” minutes.


Notes on my findings - an 11GB image is small for a Windows image. Typically Windows images for actual production deployment are larger, generally from 20GB to 30GB (in 2016). I’ve seen people deploy images as large as 160GB though. The larger the images are, the more ZSTD will pull ahead of the others and have a clear-cut superior performance. All research on it shows this, and big images in FOG will show it, too.