All things related to multicast

[Pekka Savola <>]

On Fri, 21 Sep 2007, Yuri Kolomiyets wrote:
> Pekka Savola wrote:
> > HP is sending the Cisco an (S,G, rpt) prune i.e. "I don't want that to 
> > receive back from RP the transmission from sources that we already see 
> > directly" -- and Cisco as a consequence deletes the state.
> >
> > The next PIM join however has S-bit (shortest path tree set).
> That's exactly what I was thinking. Seems like HP tries to switch from RP 
> tree to SP tree. The only problem is that RP tree and SP tree are the same 
> since traffic goes through RP because of topology, not because tree is built 
> that way. So somehow HP can't understand it and keeps trying to switch over 
> and over again.

Later you mention that with SPT switch disabled on HP, stream stops 
working at all.

In your topology I suppose RPT-only should work just fine and finding out 
why that doesn't work is probably easiest to debug and get fixed (first).

Do you mean that it works for a while and then stops (the first timeout) 
or that it doesn't work to begin with?

If the latter, the problem might also be in Cisco and you might want to 
check out that RPT state exists on Cisco and it's sending out the traffic 
to HP on the RPT state.  If it does do that, then I guess HP just doesn't 
know to handle the traffic/RPT state and the ball is clearly on their 
court?

If the former, it's likely that the problems are in HP.  There are 
probably many, the first of which is why is RPT join failing to refresh in 
a timely manner.  As Marshall hinted, one may have to check out if IGMP 
membership reports / timeouts coincide with that (i.e., if the DR loses 
IGMP state, it'll prune or not refresh the state), though I suppose that 
shouldn't be a problem in this particular case.

Pekka