Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

VMWare Backup

Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 4

Speed drops from 30 MB/s to 4-5 MB/s if vcbmounter is launched by netbackup.

If I launch vcbmounter via command line, I reach 60-80 Mb/s.

Anybody knows why this difference?

These are the command line launched

"C:\Program Files\VMware\VMware Consolidated Backup Framework


vcbMounter.exe" -h virtualcenter.domain:443 -u "user" -p "password" -c "C:\Program
Files\VMware\VMware Consolidated Backup Framework
vmName.cache" -a ipaddr:vm.domain -t fullvm -L 6 -r "M:\mnt\vm.domain" -M 1 -F 0 -m san <---
Lanched by Netbackup

"C:\Program Files\VMware\VMware Consolidated Backup Framework


vcbMounter.exe -h virtualcenter.domain -u user -p password -a ipaddr:vm.domain -r
m:\mnt\vm.domain-fullVM -t fullvm" <--- Launched by Me via CLI

http://communities.vmware.com/thread/190359

http://www.symantec.com/connect/forums/vmware-backups-are-too-slow

http://www.symantec.com/connect/forums/vmware-esx-vcb-alternatives

Here's my fustrations with NB, VCB so far.

We have been finding limitations with NetBackup for VMware. Getting round RDM's and
VMFS slowness.

Alternative to VCB on VM's with RDM's


---------------------------------------------------
We have resorted to using backups in the VM for these.
We have SAP databases and archive logs to deal with that are on RDM's. They are not too large
in size. Largest db is 240GB.
However, we are planning to put our production SAP instance, 1TB, soon onto the VM
infrastructure.

Yes, doing backups in the VM does impact ESX host I/O and others sharing resources, but we
are a relatively small VMware shop, and as yet has not produced any adverse effects.
We have added a second NIC to most VM's and assigned that to a BACKUP VLAN in VMware
Infrastructure. The physical master server is also attached to this BACKUP VLAN.

SAP's brbackup tools direct db backups/archives to a large NFS mounted staging area on our
master server over the BACKUP VLAN.

From there we perform local NetBackup filesystem backups of previous backups, leaving the
recent ones on disk.

For the VM's we have RDM's attached we perform FullVM backups of there VMFS vmdk disk
which holds the OS. RDM's are of course ignored as they are unable to be snapshotted.

VM's with RDM's attached are Linux.

The above is not ideal, as you want to perform all backups in a Virtual environment for the most
LAN free, and impact free.

As you suggested there is snapping at the SAN level, if you have the option available to you.

Medium to Large size VMFS LUN VCB backups slow (>50GB)


------------------------------------------------------------------------------------
The speed/performance aspect of VCB backups of VM's that have a VMFS LUN size over 50GB
and beyond when performing a FULLVM is poor.
The performance tails off dramatically and I mean dramatically during the snapshot copy to the
staging area. This seems to occur round about the 50GB mark of a vmdk file. Up to that point the
performance is as expected.
My SAN design is not the issue here.
I have tested throughput using the vcbmounter performance tests (as per the George Winter
Symantec WhitePaper) on the proxy doing a fullvm (which BTW uses the 2GB chunk method)
[see below]
Results were astonishing when doing this; then came the shocker when using NetBackup in front
of it. (which doesn't allow the 2GB chunk method option)

vcbmounter
-----------------
VM (Windows 2003 Server) is on a 270GB LUN via a 4Gb Fibre Channel connection to SAN
with 15K 300GB disks.

The VM consists of 3x VMDK files - 12GB, 15GB, 220GB = 247GB


Note: these files are fully provisioned VMDK on disk - ie fat not thinned.
Total Size Backed Up:    247GB

Time to completion of the snapshot process to the proxy staging area:


Total Time Taken:        44 minutes.

Thats over 5.5GB/min throughput. Fantastic.


NetBackup for VMware
-------------------------------

Time taken: 16hrs 3mins !

96% degradation in performance.

References:
Backing Up VMware with Veritas NetBackup | George Winter | January 2009
https://www-secure.symantec.com/connect/articles/u...

On Page 17

• Once a VMware Consolidated Backup snapshot is created, data is transferred from the VM
datastore to the backup proxy mount point. The completion speed of the snapshot process can
be significantly enhanced if care is made to ensure that the data path from the datastore to the
snapshot mount point (that is, staging area) is as fast as possible. The snapshot mount point
should be configured over as many dedicated spindles as possible. While we recommend that the
snapshot mount point be configured as fast disk, it does not need to be highly available. The
snapshot mount point is only used temporarily during the backup process and is not a permanent
repository of backed up data. The vcbMounter.exe command can be used to perform snapshot
creation and transfer performance tests. This can be done by invoking the vcbMounter.exe
command from the backup proxy. For example, if you wanted to test the snapshot throughput
rate
of a VM named vm100.veritas.com to a backup proxy named proxy1.veritas.com, you would run
the vcbMounter.exe command from the backup proxy as follows:

vcbMounter.exe  -h esx1.veritas.com  –u root  -p foobar  -a


ipaddr:vm100.veritas.com   -r d:\mnt\vm100.veritas.com  -t fullvm

The above does not give a true representation of what you expect in NetBackup for VMware as
whats not shown here is that a default is applied here which is the 2GB chunking of your vmdk
file onto the proxy staging area.

Thanks for the heads up on the 6.5.4 EEB for the monolithic option.
(You need to install 6.5.4 first then get hold of this from support)

Detailed here: http://seer.entsupport.symantec.com/docs/322769.htm


Support provided the EEB (a windows installer package), I ran it on my vcb proxy server but
complained about no existing vfm_master.conf - added a -create flag and extracted the files.

This EEB contains an update to the vfm_master.conf file and the libfi_VMware.dll

Well my proxy does not have the vfm_master.conf file only a vfm.conf file, so when it extracted,
I copied over and replaced the file on my (Linux) master server.

Now after restarting my NB Admin Console you get the following option in the policy.

Setting the monolithic option to 0 splits the VMDK's copied to the staging area in 2GB chunks
and is way quicker than have been experiencing.

You might also like