How fast NetBackup 7.6

[konmaninfas]

Hi All,

I have two ESX5.5 server and a Master server(Windows 2012 R2) connected 10Gbps LAN.

Then I heard that NetBackup uses neither File transfer nor Block transfer, but Buffer transfer.

So, I wonder how fast Buffer transfer compared with File transfer or Block transfer.

Would you tell me about that if there is any information?

[Marianne]

NetBackup is only as fast as your infrastructure... 
Disk/LUN layout, SAN, network, backup devices, etc....

Have a look at this benchmark test where NBU was tested against 2 competitors with the same infrastructure:

http://www.symantec.com/enterprise-backup-software-comparison-benchmark/

[Nicolai]

Netbackup uses exactly what VMware offers of connectivity - nothing more or less.

Can you possible show us us links about "File transfer nor Block transfer, but Buffer transfer" ?

[sdo]

To answer... How fast is NetBackup... I have seen...

...one pair of Windows based HP DL360 NetBackup Media servers (i.e. just two servers) to ingest VMware vSphere VMs via FC SAN based VADP from FC SAN based VMFS datastores... achieving sustained 800 MB/s from FC SAN HDS VSP plus 400 MB/s from FC SAN NetApp (and yes that is MB/s and not Mb/s) - so that's 1200 MB/s - all via FC SAN - at the same time - moving through this pair of Windows based media servers - and being sent via cross-site (metropolitan area) dual fabric trunked long-distance (circa 30 km) SAN ISLs to a FC based VTL target at a remote site (site B).

...the limiting factor on this test was that the FC SAN based VTL was unable to ingest more than 1200 MB/s (i.e circa 1.2 GB/s).

And then in real world usage...

...the same total bandwidth achieved... which was a mixture of block based VMware VADP backups (i.e. the above two Windows based media severs) plus well over a hundred other live active file system and database agent LAN based backups being processed by some other Solaris based media servers at the same site - again saving to same FC VTL - and all of this was Site A backup clients being saved across site to site B - AND - the same type and same volume of backup activity from Site B to Site A.  Finally on top of all this were some FC SAN based duplications from the same FC SAN VTLs (at each site) to local site FC SAN attached physical LTO tape, by the Solaris media servers (i.e. the Windows based media servers were not used to duplicate from VTL to tape - and were purely for VMware FC SAN based VADP backups).

All controlled by one (stretch clustered) VCS and Solaris based master server.

I dare say the whole environment could have moved more data and more quickly, but the VTLs were a combined bottleneck of 2400 MB/s (c. 2.4 GB/s).

.

So, Marianne is right (again!), NetBackup is only as fast as your infrastructure.

[sdo · Accepted solution]

Update:  I should add that all of the above was back in v7.1.0.x and v7.5.0.x, and before client side de-dupe and accelerator really speeded things up.  The only optimizations used above were BLIB/CBT on the VMware backups and some hardware compression cards in the VTLs, and embedded LTO tape compression.

.

As for more modern configurations, then de-dupe really does mask actually physical transfer rates.

I recently did another test, one server's first ever full backup (accelerator + change journal + client side de-dupe) of a quiet (low rate of change) VM based file server (just holding binaries, kits and manuals etc) of 890 GB took 3.25 hours and de-dupe saved circa 60% (so the actual data moved was only around 40% of 890GB) - and the next full (accelerator + change journal + client side de-dupe) backup of 890 GB took 2.5 minutes (i.e. an apparent transfer rate of around 8 GB/s), and NetBackup AIR replication took 2.5 minutes too.