5 Steps to Dedup Performance Testing with HP StoreOnce VSA



VSA Performance

 

 

 

 

 

5 Steps to Dedup Performance Testing with HP StoreOnce VSA

Now that I have the HP StoreOnce VSA installed in my Lab using the instructions from my last post, it is time to see how well the target- and source-based reduplication works.  Target for storage savings and source for faster backup and lower network utilization.  For these tests, I decided to use Data Protector 8.1 with Catalyst.  Symantec Netbackup with OST could also be used, but I am surrounded by great Data Protector consultants, and I happen to be fond of Data Protector, so I chose DP 8.1.  My goals of this testing were to see:
  1. If a Catalyst store on the target StoreOnce VSA yielded different performance or deduplication results than CIFS share store on the target StoreOnce VSA.
  2. If Data Protector source-based deduplication to a Catalyst store could reduce the backup window for a ROBO client on a slower network.
Here were the steps I took after installing the HP StoreOnce VSA:

1) Install Data Protector 8.1 Cell Manager

  • Download Data Protector trial from here
  • Accept the EULA
  • EULA










  • Choose your supported OS.  I chose Windows Server, but HP-UX and Linux are also supported.
Download






  • Mount the ISO and install.
Setup














Welcome



















  • Make sure you choose StoreOnce Software Deduplication everywhere (source-based dedup)

Options











 

  • Install Data Protector Client on a Remote Machine
    • For space reasons, I chose a non-graphical Linux VM on my laptop.  RedHat 6 to be exact as my company has a subscription to RedHat.  One of the nice features of Data Protector is that client software is included at no extra charge.  I downloaded and mounted the HP-UX ISO on linux (mount /dev/cdrom /media/cdrom) and ran “sudo /media/cdrom/LOCAL_INSTALL/omnisteup.sh”.  I chose to install the media agent, the disk agent and the StoreOnce Software Deduplication options.   Make sure the prerequisite xinetd is installed first (yum install xinetd).

 

  • Import the client into the Cell Manager
    • If I had taken the time to create a Linux installation server, I could have pushed the client software and registered all at the same time.  I could have easily pushed and registered a Windows client from the Windows Cell Manager, but I did not want to create another large Windows VM on my laptop for this test.

 

2) Create a Non-Source-Dedup Catalyst Backup Job


  • First, define the StoreOnce as backup device
    • DP Device

 







  • Every StoreOnce Catalyst device needs a Catalyst gateway running on a server OS.  The type of GW OS does not have to be the same as the Client OS. The Cell Manager makes an excellent gateway.  My Cell Manager is named vCenter-W2K12.















 


  • Create or Choose a Catalyst Store
    • When you click the “Select/Create Store” button in Data Protector, you can actually create store if you need to.  Nice integration.

 























  • Creating the Catalyst store in the StoreOnce interface ahead of time is just as easy.  Just click Create and define the store.



















  • Now let’s go create the backup job to write to Catalyst.  For this example, I used a weekly full.
  • Choose Backup from the DP drop down
  • Right click on Filesystem and choose Weekly_Full.  Do not click “Source-side deduplication” for this backup test.  Click OK.




















  • Choose the system that you want to back up and a directory to back up.  I created a quick Linux VM on my laptop as my ROBO client and scp-ed 3.68 GB of random downloaded file types from my laptop to it.

Linux Client






Backup Spec






















  • Click Next
  • Choose the Catalyst destination you created and, optionally, a mirrored destination.  Click Next.
Destination








  • Change the description if you like and then click Advanced.
Backup Options
















  • In the Other tab, choose “Display statistical info” for better backup performance data and click OK and Next














  • Accept the default schedule and click Next
Schedule


















  • Review the summary and click Next

Summary








  • Choose Save as… and give you backup a name

Save As















Save As 2






















  • To run the job now, right click the job and choose Start Backup...
Start Backup














  • Keep the defaults of Full and High and click OK.  One of the nice features of deduplication is that we can run a full backup every time.

Start Options














  • The backup will start.  When the backup completes, you will see statistics that look something like this:
2604:0:31617084469562144]@vcenter-w2k12": 
Mbytes Total: .................      6445
MB Mbytes Written to Disk: .......        77 MB 
Deduplication Ratio: ..........      83.7 : 1 
[Normal] From: VBDA@dfaucher-rh6 "/"  Time: 5/6/2014 8:41:25 PM
Backup Profile:
Run Time ........... 0:25:39

Backup Speed ....... 4.06 MB/s

  • Make note of the Run Time and Backup Speed

3) Create a Source-Dedup Catalyst Backup Job

  • Follow the same instructions in 2) for creating a backup job, but this time, check the Source-side deduplication check box
Source Dedup





















  • Run your backup and make note of the Run Time and Backup Speed

4) Create Non-Source-Dedup CIFS Backup Job

  • For this test, create a CIFS (NAS) store on your StoreOnce, define it as a backup device in Data Protector and use that store as the backup device
CIFS









  • The Data Protector device type will be File Library and the Repository will be the UNC path to the share.

File Library














Share Name







CIFS Backup









 

  • You will not be able to choose Source-side deduplication in Options as your destination is not a Catalyst store.
  • Run your backup and make note of the Run Time and Backup Speed


5) Compare the results.  Here are the results of my three runs backing up the same Linux client on my wireless laptop to a StoreOnce VSA on my vSphere server.

  • Run WITHOUT Source-Based Dedup to Catalyst store (Base for comparison)
    • Mbytes Total: .................      6445 MB
    • Mbytes Written to Disk: .......77 MB
    • Deduplication Ratio: ..........  83.7 : 1
    • Run Time ........... 0:45:03
    • Backup Speed ....... 2.31 MB/s
  • Run WITHOUT Source-Based Dedup WITHOUT Catalyst (CIFS. About the same speed, but 1/3 the target dedup)
    • Mbytes Total .................    6267 MB
    • Run Time ........... 0:39:05
    • Backup Speed ....... 2.67 MB/s
 
  • Run WITH Catalyst Source-Based Dedup to Catalyst store (5.6X faster, 90:1 dedup)
    • Mbytes Total: ................. 6447 MB
    • Mbytes Written to Disk: .......71 MB
    • Deduplication Ratio: .......... 90.8 : 1
    • Run Time ........... 0:08:03
    • Backup Speed ....... 12.95 MB/s

Bar Chart















So to answer my original two questions:

  1. If a Catalyst store on the target StoreOnce VSA yielded different performance or deduplication results than CIFS share store on the target StoreOnce VSA. Minimal speed difference, but Catalyst target dedup yielded 3X dedup over CIFS.
  2. If Data Protector source-based deduplication to a Catalyst store could reduce the backup window for a ROBO client on a slower network.  Significant backup performance increase for ROBO clients with a 5.6x improvement in backup speed.

I hope you have found this post helpful.  Feel free to contact me with any comments, questions or corrections.

Comments