Connecting Cisco Switch Stack to Xenserver Management port

[Michael Wright1709160875]

Hiya 

 

We have several Xenservers in a pool and we have connected the bonded storage ports to each of our Cisco switches that are in a stack(only 2).

 

We have also connected our management ports from the Xenservers to the Cisco switches. These are not bonded and connect to Access ports on the Cisco switch.

For some reason the interfaces on the Cisco keep going into a err-disabled state which are connected to these management ports. We are just using Access ports configured to use the native VLAN.

The err-disabled is being caused by loopbacks being detected on these interfaces.

 

How are the Xenservers constructed, could they be causing a loop, is there any special config needed?

 

Mike

[Alan Lantz]

The XenServer is a switch so if you have multiple physical interfaces going to a switch stack on the same vlan 

you will have looping issues. I would put the management interfaces on a separate VLAN.

 

--Alan--

 

[Michael Wright1709160875]

Hiya

 

We have management using vlan 1 (own subnet) 1nic

we have storage using vlan 1 (own subnet)  2 nics bonded (non etherchannel)

data using vlan 1 &50 (trunk) 1 nic

 

These all go to the same switch stack.

 

So to prevent loops we would need:

-each type of traffic on separate vlan

-would we need to use etherchannel for the storage bond? 

-as data uses 1 & 50 would other traffic not have to use either vlan.

 

Mike

[Alan Lantz]

Yes, keep those three types of traffic on separate VLAN's. Preferably yo would want your

storage on separate switches. You don't have to use etherchannel for storage, but I 

personally prefer LACP. 

 

--Alan--

 

[Tobias Kreidl]

What storage: iSCSI or NFS or both?

[Alan Lantz]

I assumed since he mentioned bonded storage he was talking NFS. 

 

--Alan--

 

[Tobias Kreidl]

Makes sense, yes.

[Michael Wright1709160875]

Yes NFS storage. Where not using etherchannel for the bond. Its the other type of bond, citrix docs says when not using etherchannel for the bond it is still recommended to use a stack.

can you confirm.

[Tobias Kreidl]

Any iSCSI connections should always be on separate, unbonded, dedicated NICs. If you can use two separate switches that's good from a high availability viewpoint. Since not a bond, I'm not sure any sort of stacking is needed. We just use private nets, like 10 or 192.168 for iSCSI since packets don't have to go anywhere else other than between the server and storage device.

 

-=Tobias

[Alan Lantz]

For bonding you can active/passive, active/active, or LACP.  As mentioned earlier I prefer LACP. It does require you

do so some switch configuration, but the upside is I find its more resilient and has lower failover times compared

to active/active.  If you have separate storage switches you can also use jumbo frames to squeeze more out of 

your bandwidth. I wouldn't use active/passive unless you don't want to do switch configurations for LACP and

your switches have trouble with active/active (some do).

 

--Alan--

 

 

 

[Tobias Kreidl]

Agreed, for NFS: use bonds, preferably LACP, if supported on your switches. If you split between switches, they must be stacked.

 

-=Tobias