LTO

How to create an independent backup network

How to create an independent backup network

Today we look at independent backup networks especially in regards to LTO 7 and VMware ESX hosts. Be aware – this very example also applies to any backup to disk (B2D / Backup-2-Disk) solution. But a good reseller / vendor would inform you about this right away anyways.

LTO 7 and later like LTO 8 drives have a write speed faster then a 1 GBit network can handle, making it now really necessary to think about options. On top of it, you do not want to over utilize the LAN side of your servers so that the impact on the user / application facing side stays minimal. This leaves you with two options, you can group switch ports assuming you have enough 1 GB ports and use them, you will need at least 3 ports combined, or you create a whole backup network on a 10 GB basis.

Let’s run some numbers:

  • LTO7 has a write speed of about 300 MB/s uncrompressed and up to 750 MB/s compressed
  • LTO8 (L8) has a write speed of about 360 MB/s uncrompressed and up to 900 MB/s compressed

Now – your network connection is meassured in MBit/s not MByte/s. Byte to bit is 8 bit are one byte, so we need to multiply those speeds in byte with 8 bit too see the network speed numbers.

  • LTO7 uncrompressed = 300MB/s * 8 = 2400 MBit/s
  • LTO7 compressed = 750MB/s * 8 = 6000 MBit/s
  • LTO8 uncrompressed = 360MB/s * 8 = 2880 MBit/s
  • LTO8 compressed = 900MB/s * 8 = 7200 MBit/s

Assuming you want to go with grouped ports, you see that with LTO7 you would need 6 ports and LTO8 7 to 8 ports to fully utilize the speed and minimize your backup window. Additionally think about the read speed that might affect you as well – not just for recovery but for the verify of your backup.

Now – this means – add at least one 10 GB switch and one 10 GB NIC to each server – let’s do this with an example:

  • 3x VMware ESX hosts – LAN side and management is configured 1 GB – we assume there is some kind of storage behind them that has the iOPS and speed we need like an SSD based storage
  • 1x Backup media server that has an LTO7 or LTO8 drive connected – 1 GB on the LAN side

What we need – minimal:

  • 4x 10 GB NICs
  • 1x 10 GB switch
  • 4x CAT6e or CAT7 cables

What I would recommend – nice to have:

  • 4x 10 GB NICs – dual port
  • 2x 10 GB switches
  • 10x CAT7 cables – 2x to stack/trunk the switches if not stacked otherwise

This is a nice to have – a fail-over, but the minimal configuration is sufficient as well.

Cable this all in – create a new IP-scope / VLAN on the backup side – you do not need any default Gateway etc. on the Backup-Network side (10 GB). Just an independent IP scope and have every host assigned a static address.

This keeps the regular network traffic and any broadcasts away from this network and your backup will run totally independent. You might need to disable your anti-virus solution on this NIC / IP-scope on the backup media server as well, cause it might actually influence the speed quite drastically. Having it separated and independent helps keeping the security up.

On the VMware hosts – I like to even allow VMware to vMotion on this backup-LAN – simply because it is extremely efficient there – independent from your LAN and if you have it from your iSCSI network as well. But that’s just an idea.

Now – the backup – how will it grab the data from the 10 GB side of your VMware hosts – especially if you have a vSphere cluster and grab the backup through the cluster?

Simple – you adjust the hosts file on your media server. Each and every VMware host needs to be listed in the hosts-file on the media server with the IP that it has in your 10 GB backup network. This way DNS and everything will act normal in your environment, only the backup-media server will reach out to those hosts on the 10 GB network due to the IP resolution of those hosts. This is the easiest way to accomplish this.

You will not to add a 10 GB connection, backup-network IP address etc. to your VMware vSphere controller – it can stay on your LAN or server-management network as is. This also means there is no reason to mention him in the hosts file on the media server.

How this works:
your backup will contact the vSphere controller on the LAN side
it will then be redirected to the host that currently holds the VM you want to backup
the media server now will contact the VMware host directly – due to the hosts-file entry on the 10 GB backup-network
backup will process…

This of course would work with a physical server as well – like a physical file-server etc. – though, today this is rather rare and especially VMware backups are actually large files that benefit most from the LTO7 write speed so the above makes sense there most. It wouldn’t matter if you do the same to an Hyper-V environment or any other VM host/guest solution. In theory it should always work the same.

What real world write speeds can you expect?
This is the big question – here are some real world examples of this – those are single jobs on per VM basis, meaning it includes even the tape-load and tape-unload processing time and udpating the catalogs while using Veritas Backup Exec.

Backup size (VM size)elapsed time in minutesjob rate write/overalljob rate verify
4 TB6:4717,227 MB/min26,404 MB/min
2 TB6:476,822 MB/min22,233 MB/min
1.21 TB3:498,271 MB/min20,235 MB/min
147 GB0:336,491 MB/min22,655 MB/min
138 GB0:1718,403 MB/min27,726 MB/min
25 GB0:109,172 MB/min20,700 MB/min

The above list is just an example – realistically we see speeds between about 3,000 MB/min to 18,000 MB/min as for the overall speed. This is due to the VM itself for some part – thin or thick provisioned, what kind of data is it holding, how busy is the host cause we might double trouble him due to multiple drives doing backups at the same time to the same host etc… In average we see around 8,000 to 9,000 MB/min in speed, what is still great – and I wanted to show as well that it can vary quite a bit so don’t be scared. We still did improve the time the backup took from going from an LTO4 LAN based backup scenario to an LTO7 independent backup network while cutting the time in half, actually, even less then half. The slowest speeds we see today are due to systems that can only be backed up on the LAN side, while the ports are grouped there but we still don’t have the same speed as we see on the backup-network side. Many factors come in play but that all depends on the individual situation.

Hoping the information above helps some of you out there – keep in mind that your situation might be different, run some examples and ideas and if you have questions, reach out – this remains an example of what I really implemented at a company and how it affected the backup configuration and management.