huge packet loss after bond interface setup

[jurekk]

Hi, i have problem with bond setup. I have followed this video http://pve.proxmox.com/wiki/Bond_configuration_(Video) my /etc/network/interfaces:auto lo
iface lo inet loopback

iface eth0 inet manual
iface eth1 inet manual
auto bond0
iface bond0 inet manual
slaves eth0 eth1
bond_miimon 100
bond_mode balance-alb
auto vmbr0
iface vmbr0 inet static
address 10.64.10.7
netmask 255.255.255.224
gateway 10.64.10.1
bridge_ports bond0
bridge_stp off
bridge_fd 0
After reboot which applies pending changes i have approximately 75% packet loss on test CT. Any suggestions? Thank you very much.

 

[osobh]

What type of Network Card do you have in this server?

 

[spirit]

Hi, i have problem with bond setup. I have followed this video http://pve.proxmox.com/wiki/Bond_configuration_(Video) my /etc/network/interfaces:auto lo
iface lo inet loopback

iface eth0 inet manual
iface eth1 inet manual
auto bond0
iface bond0 inet manual
slaves eth0 eth1
bond_miimon 100
bond_mode balance-alb
auto vmbr0
iface vmbr0 inet static
address 10.64.10.7
netmask 255.255.255.224
gateway 10.64.10.1
bridge_ports bond0
bridge_stp off
bridge_fd 0
View attachment 1337After reboot which applies pending changes i have approximately 75% packet loss on test CT. Any suggestions? Thank you very much.

from my experience, don't use balance-alb, it's pretty much a hack with mac address tricks and it's not enough stable

the only mode which works really great is active-backup or lacp

 

[jurekk]

I have Intel® 82576 Dual-Port Gigabit Ethernet Controller

 

[osobh]

I have Intel® 82576 Dual-Port Gigabit Ethernet Controller

I have multiple servers with the same card and I found that balance-TLB and LACP work fine for me. LACP will need you to setup your switch first (bond ports), then setup proxmox in LACP mode.

 

[jurekk]

This is not what I wanted to hear so if i need really functional receive/transmit balancing with failover i need to purchase switch which supports 802.3ad. Then setup trunking on it and enable 802.3.ad in proxmox

 

[mir]

Have you tried balance-rr and balance-xor?

 

[jurekk]

Yes i tried to setup balance-rr but with same result approximately 90% packet loss. I have read that The balance-rr, balance-xor and broadcast modes need switch ports grouped together. So those modes are not switch independent. I have unmanaged switch.

 

[mir]

No, I have used both balance-rr and balance-xor on an unmanaged switch

 

[jurekk]

also balance-xor is not working correctly - 15% packet loss

 

[mir]

What brand is your switch? To me it sounds like a broken or badly behaving switch.

 

[jurekk]

LACP provides load balancing and failover at same time? For example i have topology like this (see picture). I need two things: 1. aggregate two nics to get 2Gbps throughput 2. HA when one of the switches fails. How can i achieve this?

 

[spirit]

LACP provides load balancing and failover at same time? For example i have topology like this (see picture). I need two things: 1. aggregate two nics to get 2Gbps throughput 2. HA when one of the switches fails. How can i achieve this?View attachment 1357

lacp provide loadbalancing, but by sourceip-sourceport-destinationip-destinationport connection.

So you can't get 2gbit/s from 1 tcp connection.

 

[jurekk]

Thanks for responding, I take note of it but it's not exactly what I asked

 

[Mr.Embedded]

This is a super old thread but in the 5.4 version of Proxmox I had a weird issue with LACP bonds giving me 50% packet loss. My cluster didn't come back in a stable way after a power outage. Generally I get it back with some pve-cluster restarts/massaging but this time it was a pain and I had to dig deeper. Turns out there was a config mismatch:

In the GUI (LACP) and /etc/network/interfaces = "Bond_Mode 802.3ad"

However...

/sys/class/net/bond0/bonding/mode = "balance-rr 0"

That was the weirdest thing. When I setup the servers I put a standard /etc/networking/interfaces file I always use and rebooted like I always have many times in the past and this seemed to have caused the condition mismatch. To fix I needed to change the GUI to something other than LACP, save (it was grey and I could save because it was already set how it should be) and then revert back to LACP and save again.

A reboot activated the configuration properly and no more packet loss. I had 2 servers with the exact same issue in a test lab.