N5K/N2K default behavior for multicast packets?

[michael.luo]

I thought N5K/N2K treats multicast traffic like broadcast by default, which means, packets should be forwarded to all ports within the same VLAN.  Was that correct or not?

Our application team has two virtual servers in the same VLAN.  The two servers communicates with each other with multicast.  Originally, I thought nothing needs to be done on the switch as the servers are in the same VLAN.  But application team said "it didn't work" and forwarded me a document http://help.globalscape.com/help/eft7-2/mergedprojects/eft/eft_ha_(active-active)_deployment.htm

Per the document, IGMP Snooping and IGMP querier need to be configured on the switch.  I thought IGMP was to make the network less noisy by forwarding multicast traffic to the needed ports only.  In another word, it's a "nice-to-have" feature other than "must have".  Right?  Why I have to enable IGMP Snooping and IGMP querier to make multicast worked in the same VLAN?

Thanks!

[Paul Chapman · Accepted solution]

Hi Michael -

In an environment with no PIM or IGMP snooping the nexus (NxK) switches will flood multicast traffic like broadcast and unknown unicast.  You should be able to easily determine if the multicast is forwarding or not simply by putting a laptop on the switch and listening to the VLAN with wireshark.

The key here is that the default configuration of the N5K, N1K, and UCS is that IGMP snooping is enabled with no querier, which means that the multicast traffic is likely to be dropped (once again verifiable via wireshark).

The good news is that the IGMP querier is extremely simple to set up. Pick an unused IP in the VLAN, then use this configuration on one or both N5Ks (Can be the same IP).

vlan configuration <vlan #>
 ip igmp snooping querier <unused IP in vlan>

NOTE: Do not use if you have PIM routing enabled on any router in that VLAN.

PSC

[michael.luo]

Thanks Paul.  What does mean "IGMP disabled" on the interface?  How is it different from IGMP snooping? 

N5K# sh ip igmp int e108/1/23
IGMP is disabled on Ethernet108/1/23

# sh ip igmp snooping vlan 450
IGMP Snooping information for vlan 450
  IGMP snooping enabled
  Lookup mode: IP
  Optimised Multicast Flood (OMF) disabled
  IGMP querier none

[Paul Chapman]

Hi -

Sorry for the delay.

The command you reference only works on L3 interfaces.  You would need PIM or some other feature which causes the interface to join or listen to multicast (OTV, possibly routing protocols) to see anything. In my example below, I've forced the interface to join a multicast group with the "ip igmp join-group" command.

switch# sh run int vlan 500
interface Vlan500
  no shutdown
  ip address 10.250.254.249/24
  ip igmp version 3
  ip igmp join-group 232.1.4.1
!
switch# sh ip igmp int vlan 500
IGMP Interfaces for VRF "default"
Vlan500, Interface status: protocol-up/link-up/admin-up
  IP address: 10.250.254.249, IP subnet: 10.250.254.0/24
  Active querier: 0.0.0.0
  Membership count: 1
  Old Membership count 0
  IGMP version: 3, host version: 0
  IGMP query interval: 125 secs, configured value: 125 secs
  IGMP max response time: 10 secs, configured value: 10 secs
  IGMP startup query interval: 31 secs, configured value: 31 secs
  IGMP startup query count: 2
  IGMP last member mrt: 1 secs
  IGMP last member query count: 2
  IGMP group timeout: 260 secs, configured value: 260 secs
  IGMP querier timeout: 255 secs, configured value: 255 secs
  IGMP unsolicited report interval: 10 secs
  IGMP robustness variable: 2, configured value: 2
  IGMP reporting for link-local groups: disabled
  IGMP interface enable refcount: 1
  IGMP interface immediate leave: disabled
  IGMP VRF name default (id 1)
  IGMP Report Policy: None
  IGMP State Limit: None
  IGMP interface statistics: (only non-zero values displayed)
    General (sent/received):
      v2-queries: 0/0, v2-reports: 1/0, v2-leaves: 0/0
    Errors:
  Interface PIM DR: No
  Interface vPC SVI: Yes
  Interface vPC CFS statistics

PSC

[michael.luo]

Paul,

One more question - if multiple switches are trunked together.  Do I apply the "ip igmp snooping querier" command to each switch?  Thanks again!

[Paul Chapman]

Hi Michael -

You technically only need 1 querier per VLAN.  You may want to configure a secondary for redundancy.

PSC

[michael.luo]

Yes, I understand that I only need 1 querier per VLAN.

What if the VLAN was trunked across multiple switches?  For example, if we have VLAN10 on three switches.  And those three switches are connected with trunk ports.  Do I need to configure the querier on each of the switches?  Or I can configure it on one switch?

[Paul Chapman]

The IGMP query frame is flooded out all ports on the VLAN.  So all ports for a given VLAN, including trunks, will receive a copy of the frame.

PSC

[michael.luo]

So if ten switches are daisy-chained together with trunk ports, we just need the querier command on one of the switches, correct?

Just for the sake of discussion, what if the querier command was configured on multiple switches and have different ip addresses?  :)

[Paul Chapman]

Not that it is a good idea to daisy chain 10 switches, but yes.  As long as you have extended the VLAN from end to end, they will all receive it, just like any other broadcast or multicast packet placed on that VLAN.

If you were to put the querier on a second N5K for redundancy, then each 5K would send a periodic query independent of the other.  Multicast receivers would hear and respond to both.