[c-nsp] Sup720, multicast bothers the CPU

[Peter Rathlev]

On Wed, 2011-03-23 at 20:55 +0100, Peter Rathlev wrote:
> Thanks. We'll try just adding "ip igmp snooping querier" to the specific
> SVI to see if this in itself would be enough. Next up we try "ip pim
> sparse-mode" on the SVI and "ip multicast-routing" in global.
> 
> I'll keep the list updated on how it went. :-)

We now tried both adding an IGMP Querier and enabling multicast-routing
and PIM. None of these things stopped the multicast traffic from hitting
the CPU.

I'm a little puzzled that we actually saw _some_ sources:

xxx#sh ip igmp membership 
[...]
 *,239.255.255.253              10.26.46.78     00:02:41 02:46 2A     Vl46
 *,239.255.255.250              10.26.46.77     00:02:43 02:46 2A     Vl46
 *,224.0.1.40                   10.26.46.254    00:02:43 02:53 2LA    Vl46
xxx#

Just not from the "interesting" groups. We tried restarting both a
receiver and a source to see if anything changed, and we never saw joins
or "*,<something>".

So a few questions more:

 1) Could "offset != 0" somehow mean that the flows cannot be hardware
    switched at all? Every single packet has offset != 0, i.e. no packet
    observed with offset == 0. We were looking at both an RP SPAN
    session and a SPAN session covering the VLAN in question.

 2) Could 224.0.0.0/24, which they use for this purpose though that's
    wrong, somehow be treated specially by a Sup720? Any chance it would
    help using 239.255.255.0/24 instead?

Thanks in advance.

-- 
Peter