Bonding on BL460 with Ethernet Blade Switches

[Martijn van der Kaaden_1]

I am trying to setup a proper bonding configuration with Novell OES 1 (Linux)
Basicly this is SLES9 SP2.

Configuring and starting the bond is no problem, but getting a proper failover to work gives me headaces.

My primary goal is to have failover in case of a link failure between an Ehternet Blade Switch and the core switches of my network.
In this case the bond can't rely on linkfailure-detecion only but must somehow check the path between both slaves.

I have read the Linux Bonding HOWTO, but I can't find proper best practices when it comes HP Blades with Ethernet Blade sitches on this subject.

In the current configuration, the bond retrieves its mac-address from one of it's slaves.
Wouldn't it be better to set a unique mac to the bond ? I really don't know.

Anybody any expirience with this ?

regards,

Martijn

[Steven E. Protter]

Shalom,

For bonding to work, you need to not have any MAC address information in the ifcfg-bond0 file.

I also recommend taking it out of the ifcfg-eth0/1 files

Beyond this best practices is the standard changes in /etc/modprobe.conf , nothing further.

alias bond0 bonding
options bond0 miimon=200 mode=active-backup


Some switch configurations will require update to permit proper failover.

SEP

[Martijn van der Kaaden_1]

Hi Steven,

I think I've have found my solution.

I always thought linkfailure detection wouldn't work and it would rely on arp monitoring.
However I ran into a post of someone in which he explains the following:

"Point 14.1 of the documentation says that arp monitoring does work in bladecenter (and even is the suggested solution). This is not true. When reading the sources, during arp monitoring each rx/tx counter increase is sufficient that the active slave stays alive. The arp monitoring is working well, as long as only one blade is attached to the bladecenter. As soon i add another blade to the bladecenter the failover does not work any more, because the rx/tx counter still increase with a unpluged wire from one switch. So the failover never can happen. The only way to fix this is to configure "Trunk Failover" on the bladecenter switch. This setting forwards the
Link status directly to the card and use of normal dead link detection is possible and works well."

So I went back to the GbE2c switches and saw the Uplink Failure Detection option (I never noticed it earlier !) Which gives the possibility to monitor a designated uplink (or trunk) and in case of a failure it can put another set of designated downlink ports offline.

So I have to use the miimon feature and disable arp monitoring on the bond.

The Cisco Blade switch should also facilitate this kind feature.

[Martijn van der Kaaden_1]

See previous post