Skip to main content

How do you bond/team NICs in Linux?


vNote42
Forum|alt.badge.img+13
  • On the path to Greatness
  • 1246 comments

This topic is not directly related to Veeam. But because of the Veeam Hardened repositories running on Linux, this is also an topic for Veeam.

Do you prefer bonding or teaming?

https://ngelinux.com/difference-between-bonding-and-teaming-in-linux/

https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/7/html/networking_guide/sec-comparison_of_network_teaming_to_bonding

With bonding, which mode do you prefer?

https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/7/html/networking_guide/overview-of-bonding-modes-and-the-required-settings-on-the-switch

I personally preferred bonding with ALB. But lately I have heard more and more about problems with this mode.

What are your experiences with this topic?

9 comments

Mildur
Forum|alt.badge.img+12
  • Influencer
  • 1035 comments
  • November 3, 2022

Questions about personal preferences should be marked as „content“ :) 

Not as a „question“:

 


mkevenaar
Forum|alt.badge.img+15
  • Veeam Vanguard
  • 148 comments
  • November 3, 2022

The method of bonding (and probably teaming as well) depends on your switch and how you have configured the switch. In my case, I have 2 switches that work together in a cluster and I use LACP (802.3ad or mode 4) for bonding This needs to be setup on both the machine and the switches. However if I could not use that. I prefer bonding mode balance-alb (mode 6).

 

If you are running virtual machines, bonding IMO does not make any sense, redundancy should be configured on the hypervisor level

 

I have never setup teaming on Linux, so I cannot comment on that


vNote42
Forum|alt.badge.img+13
  • Author
  • On the path to Greatness
  • 1246 comments
  • November 3, 2022
mkevenaar wrote:

The method of bonding (and probably teaming as well) depends on your switch and how you have configured the switch. In my case, I have 2 switches that work together in a cluster and I use LACP (802.3ad or mode 4) for bonding This needs to be setup on both the machine and the switches. However if I could not use that. I prefer bonding mode balance-alb (mode 6).

 

If you are running virtual machines, bonding IMO does not make any sense, redundancy should be configured on the hypervisor level

 

I have never setup teaming on Linux, so I cannot comment on that

Thanks for your comment, Maurice!

I prefer ALB to LACP because ALB is switch-independent. Therefore it is simpler to set it up. And when it comes to a disaster recovery it could be necessary to set up new switches as well. With ALB all I need are uplinks in the right VLAN.


vNote42
Forum|alt.badge.img+13
  • Author
  • On the path to Greatness
  • 1246 comments
  • November 3, 2022
Mildur wrote:

Questions about personal preferences should be marked as „content“ :) 

Not as a „question“:

 

you are right, both my post for today are in the wrong category … am a little out of practice

@Madi.Cristil could you please move my posts?


Madi.Cristil
Forum|alt.badge.img+8
  • Community Manager
  • 613 comments
  • November 3, 2022
vNote42 wrote:
Mildur wrote:

Questions about personal preferences should be marked as „content“ :) 

Not as a „question“:

 

you are right, both my post for today are in the wrong category … am a little out of practice

@Madi.Cristil could you please move my posts?

Hey @vNote42 :)No worries! With the new job, I bet there's a lot to learn! :D Already transformed them in conversations! 


mkevenaar
Forum|alt.badge.img+15
  • Veeam Vanguard
  • 148 comments
  • November 3, 2022
vNote42 wrote:
mkevenaar wrote:

The method of bonding (and probably teaming as well) depends on your switch and how you have configured the switch. In my case, I have 2 switches that work together in a cluster and I use LACP (802.3ad or mode 4) for bonding This needs to be setup on both the machine and the switches. However if I could not use that. I prefer bonding mode balance-alb (mode 6).

 

If you are running virtual machines, bonding IMO does not make any sense, redundancy should be configured on the hypervisor level

 

I have never setup teaming on Linux, so I cannot comment on that

Thanks for your comment, Maurice!

I prefer ALB to LACP because ALB is switch-independent. Therefore it is simpler to set it up. And when it comes to a disaster recovery it could be necessary to set up new switches as well. With ALB all I need are uplinks in the right VLAN.

You can backup your (managed) switch config with a tool like Rancid or oxidized. If you cannot setup your new switch with LACP, disable one interface (shut interface for example), that way LACP detects there is only one interface alive and should start working as expected. Login to your server, reconfigure and done

 


BertrandFR
Forum|alt.badge.img+8
  • Influencer
  • 526 comments
  • November 3, 2022

Never used Teaming on linux, only bonding with LACP. Recently vlan tagging too.


Chris.Childerhose
Forum|alt.badge.img+21
  • Veeam Legend, Veeam Vanguard
  • 8307 comments
  • November 3, 2022

We do teaming on Linux servers for a new service we implemented.  Initially it was set up as Round Robin but then we changed to ALB I believe (I can confirm if needed).  Everything seems to work well as we need throughput on this service and are also moving to 25GB now on the servers. ðŸ˜Ž


Ralf
Forum|alt.badge.img+1
  • Comes here often
  • 79 comments
  • November 4, 2022

Just be cautious with switch independent boding modes and Cisco ACI. ACI is too damn clever to work with simple active-active bonding. That lead to some unexpected results here like admin down ports, blocked MAC addresses. In Linux we use active-backup so this was no issue, but in Windows we use the MS recommendation Switch Independent with Dynamic Load Balancing. We never had a problem with this before, but with ACI we now need to migrate all Windows servers to LACP/802.3ad.

 

https://www.ciscolive.com/c/dam/r/ciscolive/emea/docs/2020/pdf/R6BGArNQ/TECACI-2009.pdf