Join/leave multicast on 'bladeA' briefly disrupts traffic on 'bladeB'

[Thijs Van Severen]

hi all

i have a working UCS setup with multiple blade server chassis that we use for video processing tests.

We have a range of video test feeds available on the network (all multicast in SSM range) that all servers can join.  No problem there, works just fine.

However, we recently discovered that when you have 2 blades that both join the same multicast feed, they can influence each other :

On serverA and B we join (IGMP) the same test feed > both get the video feed > OK

Now when we leave the multicast stream on serverA, we get a short interruption of that feed on serverB

Both servers have a dedicated mgmt vnic an dedicated video vnic.  The video vnics on A and B both connect to vlan 1601.

It almost seems like the join on the video vlan interface is removed (even though serverB is still joined to the stream) and ten immediately re-joined.  Only a small number of packets are lost, but in our application this is a showstopper.

Any ideas ? 

[Kirk J]

Wonder if the UCSM/FIs have fast leave or explicit-tracking enabled...

What firmware level are you on?

From FI nxos:

FI-A(nxos)#show run all | in igmp

Kirk...

[Thijs Van Severen]

Hi @Kirk J 

Thanks for your reply !

Below you can find the requested info :

KOR-FI02-J09-B(nxos)# sh run all | i igmp
logging level igmp 2
ip igmp snooping
ip igmp snooping last-member-query-interval 1
ip igmp snooping report-suppression
ip igmp snooping explicit-tracking
ip igmp snooping link-local-groups-suppression
ip igmp snooping optimised-multicast-flood
ip igmp snooping querier-timeout 255
ip igmp snooping query-interval 125
ip igmp snooping query-max-response-time 10
ip igmp snooping robustness-variable 2
ip igmp event-history mtrace size small
ip igmp event-history igmp-internal size small
ip igmp event-history vrf size small
ip igmp event-history group-events size medium
ip igmp event-history group-debugs size medium
ip igmp event-history policy size small
ip igmp event-history cli size small
ip igmp event-history ha size small
ip igmp snooping
ip igmp snooping report-suppression
ip igmp snooping link-local-groups-suppression
ip igmp snooping optimise-multicast-flood
ip igmp snooping mrouter vpc-peer-link
ip igmp snooping event-history vlan size medium
ip igmp snooping event-history vlan-events size small
ip igmp snooping event-history igmp-snoop-internal size small
ip igmp snooping event-history mfdm size small
ip igmp snooping event-history mfdm-sum size small
ip igmp snooping event-history vpc size small
ip igmp snooping event-history rib size large

[Kirk J]

Hmm, seems like those settings are optimal to avoid the condition, that you are hitting.

You are probably going to need open an official tac case, get a FI port monitor session going for the uplinks handling that traffic (for wireshark purposes), as well as crank up the debug level on the FIs for igmp related.

Kirk...