Will switching to 10Gb Eternet speed-up my backups ?

[SAPNWADM]

Hello,

we are using a Repository on a Windows-Server which is connected to our ESXI Hosts via 1 Gb Ehternet. The processing speed is nearly 200 MB/s . The bottleneck is always the network . The perctenages for the components are:

Source 14%
Proxy: 23%
Network 51%
Target: 12%

In our landscape there are 4 ESXI-Servers with 4 VB&R Proxy-VMs 2 P2000G3 and 1 HP 380PG8 as Backup-Server. Will increasing the Ethernet to 10 Gb double the processing speed ? Because Network is 50% and the next limiting factor (Proxy) is 24% ?

The investment only for Backup will be under 3000 Euro for 5 GB10 Ethernet Adapter and one Netgear XS708E Switch.

Thank you

Heiko

[veremin]

Do you have a separate proxy server? Or repository server plays the proxy role, as well? Thanks.

[SAPNWADM]

On every ESXI-Server there is a Proxy-VM installed.

[veremin]

Then, indeed the network connection between proxy and repository seems to be the weakest point. However, I don’t expect that the processing speed will be doubled by increasing Ethernet to 10 GB, as the major bottleneck will be switched to a different component, and that component might become a limiting factor.

The last but not the least, I’ve to admit that even your current processing speed is quite a decent one.

Thanks.

[SAPNWADM]

Thanks for your informations. I have to backup a cluster with 140 VMs and a total size from 35,8 TB. Therefore i must stricly plan the backups and every failure with a retry of a vm will become critical. My hope was to gain some time.

Greetings

[veremin]

Of course, your backup performance should increase by improving the weakest part of the processing chain. My only doubt is that you'll get doubled performance speed. Thanks.

[dellock6]

Processing speed is already above the limit of a single GB link (125 MBs), and this can be explained by the fact that multiple proxies are sending backups at the same time to the repository. Seems indeed the repo itself could sustain a higher rate, but how is this repository connected now to the network? A single 1GB link or you are already using MPIO and multiple links?

[SAPNWADM]

Today the backup repository is connected with a 1GB Ethernet link. In the future it would bei 10 Gb ethernet like the ESXI Servers. I thought that the processingspeed is higher than the 125mb/sec because of deduplication, swap/empty spaces and compressing algorythm during proxy transfers.

[dellock6]

That too. But if you have a second link on the repository, the quickest and cheapest improvement you can try is to enable multipath on the repository and see if the processing speed and bottleneck stats will change.

[SAPNWADM]

I tried to activate a Network Team within Windows 2012 R2 with 4 Ethernetports. But i could only activate switch independent team, because the network guys won´t help me with a LAG on the switch. Therefore all incomig traffic is handled by one Ethernetport.

What can i do to activate mulipath within veeam ?

[dellock6]

One "dirty trick" could be to use two nics, use two different IP addresses, and map twice the repository using the two IP. If one IP points to a folder and the other one to another, you will have like two different repositories. By splitting jobs in two, using both repositories, you can see if the overall speed is going to increase, even without multipathing on the network side.

[SAPNWADM]

I tried something similar (I think...)

We have to seperate networks for backup (10.10.*) and for client access (172.17.*)

Therefore i installed 2 VB&R proxys in the 10.* network and 2 in the 172.17-* network. i thought, that the proxys will use the different networks for the traffic. But only the 172.17. proxys were able to hotadd the VM-volumes. I hadn´t done much research in this szenario. Perhaps i should investigate this more ?

[SAPNWADM]

Hello Dellock6,

thank you for your advise. I simply created another backuprepository like you told me ... and it works for both networks. I simply doubled my datatransfer to 2 Gbit.

[dellock6]

Nice!
Let us know the new performances you can reach now if you want.

Thanks,
Luca.

[SAPNWADM]

The configuration has an error. Both networkadapters will start the 2 jobs with nearly 1 Gbit. But after some time the network transfer breaks down and pauses on one Adapter.

I tried to transfer 2 files to the backupserver to look if the harddisk could be a problem, but both files were transfered with 110 mb/sec and 107 mb/sec at the same time.

Therefore it must be a problem with another component in the backup chain. I will do some reasearch if the proxys (or the ESXI Server) is the problem.

Greetings

Heiko

[SAPNWADM]

After playing with the configuration parameters from the ethernet adapters i have a Transferrate from 294 Mb/sec. The first job runs with 180 mb/sec, the second job with 114 mb/sec. Therefore i can handle 2 Jobs at the same time, that had run in the past for 8 and 10 hours within 12 hours today.

Big thanks from me

Heiko

[VladV]

Could you elaborate the settings you applied on your NICs that improved your speed?

Thanks

[matarvai]

When we upgraded from 1GB network to 10GB we only less then 5% increase in our backup speeds. Bottleneck just moved to another component.

[dellock6]

Thanks for the feedback, so as I supposed the network was indeed a bottleneck, but not by far the worst one. I suspect now the new bottleneck is target?

[SAPNWADM]

I changed the setting for interrupt moderation to extreme.

I think that changing to 10 Gb Ethernet would speed up our backup speed more than 5% because with the 2 backup repositorys i can achieve a processing rate 300 MB/sec. I tried 4 parallel job with 4 repositorys(for each Ethernet port) on the server, but than a problem with the storage pool from Windows 2012 R2 occurs. The was a strange problem. That d: Volume changed to offline during transfer. i think that is a filesystem load related problem. (4x125mb/sec on Harddisk could be a showstopper, i know ) I will do some research to this topic.

At the moment i change the whole settings on our test system, therefore this tweaking has no influence on our productive backups.