This one has stumped the VMware tech and Cisco Tech, so I thought I would bounce it off you gurus...
We have 6 hosts running ESXI 5.5. Each host has 3 connections to a Cisco 3750G switch stack. For performance reasons, we need to use aggregation. I set it up as per every guide out there as follows:
The interfaces (members of the Etherchannel)
switchport trunk encapsulation dot1q
switchport trunk native vlan 4
switchport trunk allowed vlan 4,66,666
switchport mode trunk
srr-queue bandwidth share 10 10 60 20
srr-queue bandwidth shape 10 0 0 0
priority-queue out
mls qos trust dscp
flowcontrol receive desired
channel-group 8 mode on
spanning-tree portfast trunk
The Ether Channel:
interface Port-channel8
description TRESX01 NIC Teaming
switchport trunk
encapsulation dot1q
switchport trunk native vlan 4
switchport trunk allowed vlan 4,66,666
switchport mode trunk
flowcontrol receive desired
spanning-tree portfast trunk
Here is the part NO ONE can figure out: When the ports are NOT a member of the ether channel, load balancing does not work as expected. When they ARE, load banlancing works GREAT.
Here is the issue: When we vmotion a machine between the hosts, the VM loses the connection to the network until either
A: some traffic is sent from the VM (like a ping) or
B: the MAC timer in the switch goes off.
You cannot ping IN, until traffic is sent OUT. I do have notify switches to on, latest v12 firmware on the switch, etc.
When we wire shark from another switch, we see the RARP packet,
The 3750s are not using them to change the MAC table. Any ideas?
