Comprehensive data protection for all workloads
Post Reply
stevil
Expert
Posts: 112
Liked: 2 times
Joined: Sep 02, 2010 2:23 pm
Full Name: Steve B
Location: Manchester, UK
Contact:

Initial backup speed on optimum hardware

Post by stevil » Dec 07, 2010 9:08 pm

As a rough guide, what speed should I be getting in network mode on an initial backup?
I'm currently running a backup and it's showing as 8MB/sec. Actual throughput seems to be averaging about 11MB/s, peaking at about 30MB/s.
Hardware is all top of the line, 3.3Ghz 6 core IBM servers, cisco Gbit switches, GB's of RAM and an IBM NAS which hosts the VM's via NFS. I've seen similar speeds on other installations we've done, but I'd always expected more.
I can't see where the bottleneck would be. Veeam is installed on a VM with 4 CPU's assigned, CPU usage is averaging 29%, which sounds very single threadish to me, not sure if that's true or not. I've done some test transfers of data across the network and got over 100MB/s.
Is there something I'm missing here, or is this speed about right given the $1m+ hardware ?
I'm using 'best' compression, since the CPU's aren't being stressed at all I thought I might as well. I'm not sure why there's a warning about using an 8-core server in this mode?

Cheers

tsightler
VP, Product Management
Posts: 5418
Liked: 2240 times
Joined: Jun 05, 2009 12:57 pm
Full Name: Tom Sightler
Contact:

Re: Initial backup speed on optimum hardware

Post by tsightler » Dec 07, 2010 10:27 pm

Network mode is very slow due to the way ESX limits throughput via the console or agent. This is especially true with ESX 4.0 (without hotfix) but it's still pretty limited with 4.1 as well. Since you're using NFS, virtual appliance mode is the way to go to get any decent speed.

Optimal compression can saturate 8 cores quite easily with just a few jobs and 8Gb FC or 10Gb iSCSI. Your seeing minimal CPU usage because your sipping data through a straw (network mode). Switch to the fire hose (SAN or Virtual Appliance modes) and you'll see some CPU's jumping.

stevil
Expert
Posts: 112
Liked: 2 times
Joined: Sep 02, 2010 2:23 pm
Full Name: Steve B
Location: Manchester, UK
Contact:

Re: Initial backup speed on optimum hardware

Post by stevil » Dec 07, 2010 11:06 pm

OK, I'll give VA mode another go. Last time I tried it performance actually dropped off compared to network mode, but I'm willing to be proven wrong. I'll set a test going.
cheers

Gostev
SVP, Product Management
Posts: 24788
Liked: 3521 times
Joined: Jan 01, 2006 1:01 am
Location: Baar, Switzerland
Contact:

Re: Initial backup speed on optimum hardware

Post by Gostev » Dec 07, 2010 11:37 pm

VA mode wins Network mode for NFS storage hands down, just give your Veeam Backup VM 4 vCPU and it should fly... network mode is specifically slow with NFS production storage for some reason.

stevil
Expert
Posts: 112
Liked: 2 times
Joined: Sep 02, 2010 2:23 pm
Full Name: Steve B
Location: Manchester, UK
Contact:

Re: Initial backup speed on optimum hardware

Post by stevil » Dec 08, 2010 8:30 am

OK, I have teh results, the VM took 3hrs (and14 seconds) to complete, that's 25MB/s. I noticed the CPU was much higher this time, around 70% on and off. So it's around 3 times faster than network mode. I did notice that it backed up "without changed block tracking" even though it was enabled, I'm not sure why that is as VA mode supports block change tracking doesn't it?

Vitaliy S.
Product Manager
Posts: 22976
Liked: 1555 times
Joined: Mar 30, 2009 9:13 am
Full Name: Vitaliy Safarov
Contact:

Re: Initial backup speed on optimum hardware

Post by Vitaliy S. » Dec 08, 2010 9:53 am

Any vStorage API mode supports change block tracking.

Gostev
SVP, Product Management
Posts: 24788
Liked: 3521 times
Joined: Jan 01, 2006 1:01 am
Location: Baar, Switzerland
Contact:

Re: Initial backup speed on optimum hardware

Post by Gostev » Dec 08, 2010 10:52 am

stevil wrote:So it's around 3 times faster than network mode.
This is exactly what everyone else reports :D 3 times.

Note that changed block tracking is not used when copying VM files, so you will only see "with changed block tracking" when the job is processing virtual disks. Also, my impression is that you are testing full backup anyway, not incremental?

Post Reply

Who is online

Users browsing this forum: Google [Bot] and 58 guests