[c-nsp] MVPN

[Phil Mayers]

Christian Meutes wrote:
>>> I just tested it in SM mode. What happens is that no more streams can be
>>> send because of the exhaustion of all data-mdt's. Switchover to data-mdt
>>> is configured to 1 and as soon as traffic arrives a data-mdt is tried
>>> to open and unfortunately fails therefore.
>>
>> Really? That's got to be a bug. What platform are you on?
> 
> What should be the behavior if no bug would be insolved? Continue to 
> stream the new group to the default-mdt?

Well, I'm beginning to doubt my own understanding now, but I was under 
the impression that the mapping of VPN group -> data group was a hash, 
and that >1 VPN group could map to a data group; so when you reach ~255 
groups, some of the data MDTs would be re-used.

http://www.cisco.com/en/US/products/ps6604/products_qanda_item09186a00800a85be.shtml

Q. What happens if the throughput of a customer's stream surpasses the 
Data MDT threshold and all Data MDTs are being used?

A. If the maximum number of Data MDTs already exists, the group 
addresses are reused. This may mean that some PEs receive CE data to 
which they have not subscribed.



> 
>>> The only solution as far as I can see is to do SSM without a data-mdt.
>>> Sure I could also use SM without a data-mdt but that would concentrate
>>> everything on the RP which would be the worst case of all.
>>
>> You seem to be mis-understanding how PIM-SM works. When a source starts
>> up, only the *first few* packets are tunneled via the RP; these few are
>> forwarded down the shared tree, which triggers the other PEs to do a join
>> to source, and the traffic then flows optimally; the RP will then send a
>> Register-Stop to the PE of the source, which stops the tunneling.
> 
> No Iam not mis-understanding how PIM-SM works :-)

Ok, sorry.

> 
> What i tried to explain was that if no more than 255 data-mdt's can be 
> created how can I stream to more than 255 multicast groups: Only 
> solution would be to do it without a data-mdt which would result in 
> forwarding everything over the Core-RP (default-mdt) with PIM-SM or to 
> do it in PIM-SSM and forward everything to every PE but at least on the 
> shortest reverse-path (SPT).

Both PIM-SM and PIM-SSM will build an SPT for both default & data MDTs.

The only difference is that in PIM-SM, the first few packets from a PE 
are sent via shared tree, which triggers the other PEs to do the PIM 
join, whereas in PIM-SSM an out-of-band mechanism (either BGP type2 RDs 
or MDT SAFI) triggers the PIM join.

Am I being clear? Basically, the RP should not see any significant load 
even in PIM-SM - a couple of packets per PE per 90 seconds as PIM null 
registers are triggered.

We have this working - 10mbit/sec IPTV groups do *not* hit the RP.