[j-nsp] P2MP LSP

Humair Ali humair.s.ali at gmail.com
Wed Jun 30 09:30:10 EDT 2010


Hey Mark

well ,

here is what I think, I have not got extensive experience in NG MVPN , I may
be wrong so feel free to contradict me, (just do it kindly ;-)

I think it depends on the size of your network ,to see which is more
appropriate, and the setup of the network.

The larger your network, the better to use selective P tunnels, the smaller
then better to use inclusive trees

The reason is,  using inclusive trees on large size network  means keeping
ressources and state accross all PE, where some of PE's potentially will be
used very few times .

In Selective , you only use the ressources of the PE's that are involved in
the MVPN.

I personally think Selective P tunnels is more scalable as it gives you more
control ,

especially if you have large requirement for IP multicast bandwith , you
decide where packets are needed, and the states are only accross these PE
routers.

This is confirms by what you wrote in your previous mail, describing the
difference between the 2 method.

But the real question is how much more state does it create , in proportion,
between inclusive P tunnels and selective , and is it worth the processing
on all PE routers compared to a slightly higher burden on few PE routers ?

I don't know how much more burden it create but I doubt it would be a
massive difference as you say.

But then of course, it is a balance of what is acceptable or not based on
one's opinions.

For me, in essence , in selective , you replace a type 1 routes by a type 3
& 4 routes.

Both use BGP to exchange information between,and if you use RSVP P2MP, then
both use the same tunnel type, only For Selective,it just add an additional
routes into the PMSI attribute during the Exchange.

Now true that selective , do also an extra distribution/mapping of C,S-C,G
between PE-PE, and has an extra policy for the type 4 but even then I am not
sure, how much more all this creates.

I am not sure in terms of forwarding state , how does inclusive and slective
compared to each other , and if it is much more for one or the other .

If anyone know, please let me know

Well that's what I think, and it might actually be worth only 2 cents , but
that's my 2 cents ;-)



On 30 June 2010 11:08, Mark Tinka <mtinka at globaltransit.net> wrote:

> On Wednesday 30 June 2010 03:38:45 am Humair Ali wrote:
>
> > i think most implementation use inclusive P-tunnels, as
> >  it easier to manage but I personnaly think it add more
> >  burden on the network.
>
> I'm positive you're already familiar with all this, but just
> for the archives:
>
> Inclusive trees eliminate the need for too much state in the
> core, but requires that all Receiver PE routers are aware
> about the Multicast infrastructure regardless of whether
> they have downstream listeners or not.
>
> Selective trees are the exact opposite. You can choose which
> Receiver PE routers will handle Multicast information so
> only interested receivers attract traffic to the right
> Receiver PE routers, but it increases the amount of state in
> the core.
>
> I would really like to see how mLDP plays into this whole
> mix.
>
> Cheers,
>
> Mark.
>


More information about the juniper-nsp mailing list