[Linux-cluster] A better understanding of multicast issues

[Corey Kovacs]

When a multicast group is "joined" the switch/router will periodically
(three mins i think) send out a query to the members to see if the
connection is still needed. If a member does not reply to this query,
then the connection is dropped for that port. If a switch is rebooted,
then it's up to the member to re-establish the connection I believe,
not the switch. Snooping is not generally a problem unless it's broken
in the switch/router firmware. If it is, then you might need an
upgrade. We use Multicast for all sorts of things and have indeed run
into some problems on devices like Flex-10 cards for HP c7000 blade
chassis that didn't do igmp snooping correctly, but we have gotten
fixes for these issues from various vendors.

If you have a planned outage for a switch, you can have your network
people relocate the querier for a particular multicast group to
another switch accessible to say a bonded pair or something. Things
get really odd if you are on two separate switches that aren't
stacked.

Generally speaking, multicast isn't hard, you just have to think backwards.

-C

On Sat, Feb 12, 2011 at 10:51 AM, Kit Gerrits <kitgerrits at gmail.com> wrote:
>
> Digimer,
>
> Did you ever get a reply from anyone?
>
> If what you say is true, failure of one of our HSRP(HA) switches/routers
> might break the cluster.
> (if they don't share multicast menberships)
>
> I would guess that  multicast groups originate in the cluster, not the
> switch.
> In that case, if the switch has been rebooted, the cluster needs to
> re-create the multicast groups on the switch.
>
> I would guess that the cluster itself needs to check if the switch is
> properly handling multicast.
> (subscribe to its own group and check if the packets are being handles
> correctly)
>
> This should provide an insight into clustering/multicast:
> http://www.cisco.com/en/US/products/hw/switches/ps708/products_tech_note0918
> 6a008059a9df.shtml
>
>
> Regards,
>
> Kit
>
>
> -----Original Message-----
> From: linux-cluster-bounces at redhat.com
> [mailto:linux-cluster-bounces at redhat.com] On Behalf Of Digimer
> Sent: maandag 24 januari 2011 16:37
> To: linux clustering
> Subject: [Linux-cluster] A better understanding of multicast issues
>
> Hi all,
>
>  It seems to me that a very good number of clustering problems end up being
> multicast and smart switch related. I know that IGMP snooping and STP are
> often the cause, and PIM can help solve it. Despite understanding this,
> though, I can't quite understand exactly *why* IGMP snooping and STP break
> things.
>
>  Reading up on them leads me to think that they should cleanly create and
> handle multicast groups, but this obviously isn't the case. When a switch
> restarts, shouldn't it send a request to clients asking to resubscribe to
> multicast groups? When corosync starts, I expect it would also send
> multicast joins.
>
>  Sorry if the question is a little vague or odd. I'm trying to get my head
> around the troubles when, on the surface, the docs seem to make the process
> of creating/managing multicast quite simple and straight forward.
>
> Thanks!
>
> --
> Digimer
> E-Mail: digimer at alteeve.com
> AN!Whitepapers: http://alteeve.com
> Node Assassin:  http://nodeassassin.org
>
> --
> Linux-cluster mailing list
> Linux-cluster at redhat.com
> https://www.redhat.com/mailman/listinfo/linux-cluster
>
> --
> Linux-cluster mailing list
> Linux-cluster at redhat.com
> https://www.redhat.com/mailman/listinfo/linux-cluster
>