[c-nsp] Nexus Layer 2 Multicast and IGMP querier

[Scott Voll]

I'm not sure if I'm understanding everything correctly, But here is my 2
cents.


Cisco Live had a BUNCH of Nexus / Multicast sessions last year because with
the whole VPC / Active / active nature of Nexus it's a different beast.

we are using n5k's with Iayer 3.

But what we have found out is that Multicast does not flow over peer
links.  Thus one of the 5k's is the source the second one can hand out the
IP mroute to another connected device without the actual traffic.

Which lead us to put all of our layer three connected devices to run point
to point l3 rather than l2 vpc's

Then we had to figure out which 5k was the source and then add IP Mroutes
to the l3 connected devices to point to the correct 5k.

To add insult to injury we also found that VMware 5.1 vswitch has major
multicast issues.  so our Multicast server is being moved to a physical
box.  At least until we can get to a nexus 1000v or upgrade VMware to a
version that doesn't have Multicast issues.

YMMV

scott


On Thu, Mar 19, 2015 at 11:12 PM, Stoward, Matt <
Matt.Stoward at team.telstra.com> wrote:

> We are in the middle of a large (1000s) of physical server (on IOS
> switches) to virtual server (on NX-OS) migration and more and more we are
> stumbling on weird situations like servers having back-to-back connections
> for heart beats where we have to make a non routable VLAN and run them out
> to all the prospective hosts to emulate this connectivity.
>
> Some applications need L2 multicast to work and because we don't have an
> any SVIs in these non routable VLANs we have been configuring up an IGMP
> querier under the VLAN configuration mode to get this to work but we are
> becoming reluctant to keep on creating more and more and more VLANs.
>
> I'm hoping to be able to run a general purpose clustering VLAN for this
> situation but I think I have a problem where one cluster might want to talk
> on say 192.168.34.0/24 and another on 10.10.10.0/24 and as the VLAN only
> allows one igmp querier address multicast will break for servers that talk
> on another range that the querier address does not belong to (e.g. igmp
> snooping querier 192.168.34.254 means that 192.168.34.0/24 cluster will
> work but a cluster talking on 10.10.10.0/24 will not work).
>
> We run a bunch of N2Ks so there is some unwillingness to turn off the
> multicast flood protection that would turn multicast into broadcast due to
> the 2K uplink cost of having lots broadcast on the network.
>
> I'm struggling to find documentation that confirms whether I can get
> different IP range IGMP working on NX-OS, and I'm hoping someone can
> confirm expected behaviour. I suspect I'm stuffed, and will need to either
> run some VLANs that turn off the multicast flood protection or commission
> even more VLANs.
>
> Regards,
> Matt
>
>
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/
>