Jump to content


Photo

Very Slow Client Machine Backup to Disk

slow backup disk network

  • Please log in to reply
3 replies to this topic

#1 ITDept42

ITDept42
  • Members
  • 2 posts

Posted 21 August 2013 - 02:52 AM

At one of our sites we are seeing backup speeds of about 30~70 MB/min on a system when another site with an exact hardware duplicate of that site is doing 1.9 GB/min on average. How can I find out what is causing this slowness and how to fix it?

Each site has a setup with 3 physical boxes, 2 of which are running ESXi 5.1, and the remaining one is running Yosemite Server Backup 10.2.00.28419. All 3 physical boxes are plugged in to the same gigabit Ethernet switch in the same rack they are all located in. We are using different software to make VM images, and are only using Yosemite to do file and system state backups.

Data is being collected by Yosemite from
  • Local RAID5 Disk of SATA2 drives ~ 170 GB in 1 hour 45mins
  • Windows 2012 RD Machine ~ 48 GB in 11 hours
  • Windows 2012 DC Machine (system state and a dozen or so small files) ~ 12.55 GB in 3 hours
Data is being saved to a Tandberg RDX removable HDD that is connected with SATA2 to the backup server. The ESXi and backup servers are not doing anything else at the time.

The only difference between the fast and slow sites in software is that faster site is using a slightly older vSphere/ESXi 5.1.0 build number.

The RD and DC machines are both hosted on 1 of the ESXi boxes which can maintain over 24 TiB/min in disk read from its RAID 10 array of 6xSAS drives, which is much more than the 71 GiB/min maximum of gigabit ethernet, which is again more than the Tandberg drive can do with its write speed of about 2.4 GiB/min

I've tried the idea of adding a null device to backup to ( https://community.ba...y-slow-backups/ ), to see if the slowness is on the backup drive. It is not. The 170GB on the local raid5 disk backs up to the null device at about 4.2 GB/min, but the RD and DC servers still have less than 100 MB/min.

A simple robocopy with the /MT switch will do about 1.9 GB/min from the RD server to the backup server. So the network can handle the traffic.

#2 Matt Orme

Matt Orme
  • Moderators
  • 463 posts

Posted 21 August 2013 - 10:51 AM

There are quite a few potential reasons for the difference in speed/throughput your seeing. In this case it doesn't sound like disk I/O based on your description of the hardware involved. It also doesn't sound like it has to do with the number of files being backed up.

Ruling the above out, and given the performance to a null device, we're left with a few remaining potential culprits.

Are the VM's at the slow site configured any differently then the VM at the faster site?
Different virtual controllers, different disk types (persistent, Thick/Thin Provisioned), Different cpu/core count, RAM,...etc.

Are there any software installations that could be limiting Yosemite's ability to push backup data to your device at the slow site?
AV, Scanning, etc.

Any differences in Firmware or drivers on the VM hosts between the two sites?

Give the volume of potential causes, I would give the above a quick review. If you do not identify any clear cause, I'd contact Barracuda support directly. You can have a case opened by sending an email to support@barracuda.com. On the off chance you're running an evaluation, support is included free of charge for the duration of your evaluation.

#3 ITDept42

ITDept42
  • Members
  • 2 posts

Posted 27 August 2013 - 05:31 PM

Thanks for the suggestions. I figured it out though. It was due to a slight difference in the Symantec endpoint protection settings between the branches, causing the slow branch RD and DC servers to perform DPI with its firewall and scan all network traffic with unchangeable settings to avoid high cpu usage by simply slowing down the network traffic.

I hate Symantec.

#4 Matt Orme

Matt Orme
  • Moderators
  • 463 posts

Posted 29 August 2013 - 10:02 AM

Glad to hear you figured it out.