no gratuitous arp N1k when second subgroup comes up again

[MICHAEL BEPPLER]

Hello,

we have an Nexus 1000v with PortChannel to no clustered upstream Switches.

Port-Profile Configuration:

config attributes:

  switchport mode trunk

  switchport trunk allowed vlan 2,6-7,64,150,607,630

  switchport trunk native vlan 1

  channel-group auto mode on sub-group cdp

  no shutdown

If one Link goes down, the VM of this Link will change to the second Link. During this change i see a maximum of one lost Packet in the Ping to this VM.

If the first Link still comes up, I see something that I don't understand.

After 6-7 seconds I have 6-7 Packets lost in the Ping to the VM that changes the Link.

The MAC of the VM will change from one Link to the other in the MAC-Address-Table on the Upstream Switches after this time.

In the Documentation I read only about gratuitous arps, when the link fails.

How can I reduce the Packet lost?

[Robert Burns]

How many links total do you have from each host to the upstream switches?   Have you configured Port Fast on the upstream interfaces as recommended to avoid any forwarding delays when the interface comes back up?

Robert

[MICHAEL BEPPLER]

I have one link from each host to each upstream switch in this port channel.

This is a test system with two ESX-Hosts and two upstream switches (C2960G).

Here the Configuration of the upstream switch Ports:

interface GigabitEthernet0/2

description esx-netz1_1

switchport trunk allowed vlan 2,6,7,64,150,607,630

switchport mode trunk

spanning-tree portfast trunk

spanning-tree bpdufilter enable

spanning-tree bpduguard enable

In every Host I have four PNICs. Two for the channel, one for vMotion and one for the service console.