[c-nsp] MVPN

[Phil Mayers]

Christian Meutes wrote:
> Hello,
> 
> is an opinion out there about sizing/designing MVPNs with more than
> 255 groups per VRF? Should I use SSM for the default-mdt and abandon
> PIM-SM for this purpose? The problem I see is the maximum cache size

I like SSM for default and data MDT, but unfortunately we had to drop it 
so we could interoperate with Junipers (which only implement the very 
new RFC, and neither of the cisco proprietary older type-2 RD or newer 
MDT BGP AF)

As far as I can see, SSM would only be of specific help if the >255 
groups were coming from >1 PE. You'd still have problems with >255 
groups on 1 PE using SSM.

> of a /24 for the data-mdt. When this limit exceeds what will exactly

Huh. I had not known that was a limit, but sure enough:

core-spare(config-vrf)#mdt data 239.192.0.0 0.0.255.255
% HASH values can not exceed 255!

> happen when all streams are still active? Streaming everything over
> the RP will produce bottlenecks and is something I want to avoid.

The traffic will not flow via the RP, since all the PEs should have 
joined towards the other PEs for the default or data group in question.

The issue is that some traffic might be sent to PEs which have no 
interest in it (if you have more active MVPN groups than data groups, or 
if it's flowing in the mdt default) but it will still flow on the source 
tree in the P-space, not the shared tree.

Whether that's actually a problem depends on the bit rate of the 
traffic, number of PEs and types of links. Not the most helpful answer 
I'm afraid.

It would be nice if Cisco would start to track the newer RFCs (hell, it 
would be nice if they'd get it out of draft and into proposed standard) 
where the "all BGP" MVPNs might help in this case.