[SOLVED] Nested Bonding: Root Bond not showing the right speed

[EmilioMoreno]

Hi all
As per my previous post, i have deployed a 2 node cluster (will become 3 as soon as possible), where each of the nodes is connected to two switches in a nested bonding configuration: https://forum.proxmox.com/threads/bonding-lag-with-multiple-switches.48398/#post-279823

Simply put Bond0 is an active/pasive bonding, formed by bond1 (LACP of 2 interfaces against one switch) and bond2 (LACP of 2 interfaces against another switch)

Both nodes are properly connected, LACP too and everything seems fine, both on physical and logical tests, but today i have seen the root bond (bond0) is showing 1Gbps speed while bond1 and bond2 are showing (as expected) 2 Gbps speed, both in the server and in the switches.

root@pve-2:~# cat /sys/class/net/bond0/speed
1000
root@pve-2:~# cat /sys/class/net/bond1/speed
2000
root@pve-2:~# cat /sys/class/net/bond2/speed
2000

By LACP mode 4 every IP may go through a different interface in the active LAG, but the doubt is if having the root bonding showing itself as 1Gbps will really limit the overall VMs/LXC to 1Gbps instead of max 2Gpbs. Anyone has an idea? If so, is it possible for bond0 to declare itself with the same speed of the active LAG?

Thanks
E.

 

[wolfgang]

Hi,

LACP is unable to increase the speed of a single connection more than a single NIC has.
You can compare it with a load balancer.
See https://en.wikipedia.org/wiki/Link_aggregation#Link_Aggregation_Control_Protocol

 

[EmilioMoreno]

Hi,

LACP is unable to increase the speed of a single connection more than a single NIC has.
You can compare it with a load balancer.
See https://en.wikipedia.org/wiki/Link_aggregation#Link_Aggregation_Control_Protocol

Hi
Thats clear, connections from a single origin ip to a single destination ip are going to pass always through the same NIC and so, they are effectively limited to 1gpbs. However if you have multiples origin IPS and multiple destination IPS, each of the pairs are going to be randomly distributed and so the overall bandwidth is increased; i am using this to "connect" non stackable switches with no issues.

However this is not what i was trying to ask, sorry if the question was not clear.

Bond1, LACP is showing 2Gbps, as per aggregation of 2 NICs on Proxmox Node and 2 Ports on the Cisco switch. I know one single ip will not be able to get more than 1Gbps, but all the traffic from multiple containers and VMs should go beyond 1Gpbs on bond1. Same for bond2

The point is when i create a new Bonding, bond0, in active/pasive, between Bond1 and bond2, and lets say bond1 is the active bond, bond0 (which is always the main interface for proxmox node) is showing 1Gpbs, despite the fact is actually referring to bond1, which is a 2gpbs

So simply put, the question is if, though bond1 can provide traffic up to 2gpbs for multiple IPs, (max per ip = 1Gbps) the fact that bond0 is shown as 1Gbps interface is going to limit the total effective speed of 2Gbps, or if there is a way for Bond0 to reflect the same bandwidth as bond1.

Is there a way to set the bond0 interface speed?

 

[wolfgang]

Is there a way to set the bond0 interface speed?

No, and it is not required.
A bond is a virtual device and it will send packages as fast as the physical devices accept them.
There is no QoS active to limit the network speed.

 

[EmilioMoreno]

Thanks!