Is Zetta Faster Than Backing up to DIsk?

A while ago, a representative from Zetta made the claim that Zetta can back up systems faster than backing up to disk.  The conversation can be found at:

http://community.spiceworks.com/topic/352281-we-can-back-up-to-the-cloud-faster-than-to-disk

 

The Challenge

 

The goal of this test is to test backup performance between Zetta and another backup-to-disk solution in a controlled environment.  To test this out, I built out a test environment in my AWS lab.  It consists of 3 identical, instances.  The stats are:

  1. AWS EC2 Medium Instance (Single-core CPU (2ECU, 3.7GB RAM)

  2. 45GB Hard Drive

  3. Windows Server 2012, fully patched, same WSUS server

  4. Same Subnet, same domain, same domain controller, same availability zone

  5. Gigabit networking including local and Internet

  6. Standard IOPS and networking profiles

  7. No antivirus/antimalware software to interfere, as this is a disposable test environment.

The backup payload consists of a single 74.7MB assets file from the video game RIFT.  I chose the file due to its slightly larger than normal yet manageable size and unusual file type.  This should prevent the backup software from running any file/type specific compression or transmission algorithms.  The file should provide a large enough transmission window to be able to time the difference between systems.  In addition to Zetta, the baseline backup solution is CrashPlan PROe, with the destination server on the same subnet using the third AWS instance.

The Test

For each backup client instance, I created an folder at c:\tobackup and set up the backup set to only back up that one folder.  I performed a backup, rebooted the server, and allowed the server to settle after booting.  From there I copied the test file into the backup folder and kicked off a backup.

The Results

The results of the test had Zetta at 1 minute, 4 seconds.  CrashPlan took 14 seconds.    That's 78% quicker.  Most of the speed came from the very quick backup job start and completion compared to Zetta, which took a bit to gain its momentum.  In prepping the test, the file copy to each test server took less than 2 seconds to complete; this is the speed one would expect to see from Robocopy or other straight-up copy to disk tools.  Sorry Zetta, you weren't faster.

Closing Comments and Observations

After the test, I ran some larger data scenarios in order to get a better feel for the performance characteristics of how each engine runs.  What I saw was Zetta grabbing the files and tossing them upstream as fast as it possibly could.  At times, I was seeing upload speeds around 43Mbps, which may have been able to spike higher, except for the CPU utilization was the bottleneck.  CrashPlan analyzes and compresses/encodes the files prior to transmission through the use of temporary files.  The end result is that it uses less bandwidth, but consumes extra I/O operations in the process.  I was seeing transfer speeds upwards of 20Mbps at times, but disk I/O was a noticeable bottleneck.  In short, consider the pros and cons of each system and how they would fit into your environment.