Help With NIC Teaming

[bmclcow]

I have a blade enclosure with many blades attached to the blade switch.  From reading the doc and experimenting, it appears that with the setting "link status only" the blades will never be aware of network problems beyond the connection between the blade and local switch.

For example, I setup NIC teaming with:

nic0 = active

nic1 = standby

By causing a failure on the nic0 path (beyond the switch), I lost connection and the standby does not kick-in.

Queston 1:

When would you ever use a standby adapter (isn;t it almost useless)?

Question 2:(I really need an answer for this one)

Assuming you use active/active instead of active/standby *and*

You have a network failure where esxi is unaware of the failure *and*

Whatever load balancing option you are using happens to try and send out on the bad path...

Will the load balancing alogrythm keep trying that bad path forever, and nver try the other path?

NB - I know thereis the "beacon probing" option but most docs ay to not use it.

Thanks

[SteveFuller2011]

Hi,

What blade enclosure are you using and which switches do they have? Many switches have a feature such as Cisco Link State Tracking that will disable the internal blade switch interfaces if all the external interfaces fail. This will mean that you should never get into a position whereby an upstream link failure will go undetected.

Regards

[bmclcow]

Thanks for the quick replies guys.

I have an HP c7000 enclosure with Cisco 3020 switches.

I think you hit the nail on the head with the link-state tracking.

It sounds similar to Virtual-Connect.  That is, there is a setting that will force the blade ports down when the uplinks go down.

Thanks again- I'll mark as solved.

[SteveFuller2011]

Glad to be of help.

You can find the Catalyst 3020 specific documentation as Understanding Layer 2 Trunk Failover. One point to note is that if the uplinks are configured as part of a port-channel the link state tracking command is applied to the port-channel interface and not the physical. This is a snip of part of the configuration we use:

!

interface GigabitEthernet0/1

  [...]

  link state group 1 downstream

!

[...]

!

interface GigabitEthernet0/16

  [...]

  link state group 1 downstream

!

interface GigabitEthernet0/17

  [...]

  channel-group 1 mode active

!

  interface GigabitEthernet0/18

  [...]

  channel-group 1 mode active

!

interface Port-channel1

  link state group 1 upstream

!

Regards

[a_p_]

Welcome to the Community,

it actually doesn't matter whether you configure active/standby or active/active for the failover in this case (I prefer active/active for load distribution). The ESXi host will only fail over in case of a link loss of the blade switch port. Beacon probing is only helpful with at least 3 uplinks, so the only really working option is Cisco's Link State Tracking, where you create a Link State group in which you define the up-  and downlinks of the switch and where the downlinks (the ones on which the blades are connected) will go down if the uplink fails.

If you don't have Cisco switches (even if you have them) you should have redundant uplinks from the blade switches to e.g. the core switch.

André