All things related to multicast

[Toerless Eckert <>]

On Tue, Mar 18, 2003 at 09:09:26AM -0500, Lappa, Joseph wrote:
> Hi,
> 
>    Here are UofPitt, we're running a fairly simple multicast environment
> 
> We want make our infrastructure more robust, but still keep it easy to
> debug.  
> With that in mind...
> 
> We're a multi-vender site with Ciscos in the core and a pair of Junipers
> for 
> our border routers.  We run HSRP in many locations with our Ciscos (8
> core 
> 6509s with MSFC running a mix of merged and hybrid code, with going to
> merged 
> soon across campus).  We want to make our multicast routing as robust
> and 
> efficient. 
> 
>    We are thinking of two possible scenarios:
> 
>       1) Make each of our core Cisco 6500s MSDP peers with each other 
>          (full mesh), with our Junipers, and configure Anycast-RP with
> PIM-SM.
> 
>       2) Make each of our core Cisco 6500s MSDP peers with each other 
>         (full mesh), with our Junipers, and run PIM BSR with PIM-SM.

If you appropriately configure your IGP for fast convergence, Anycast-RP
will provide you fast overall converence for IP multicast in case
of a router failure. The overhead of the standard Anycast-RP configuration
recommendation is that you need to configure the static anycast-rp address
on all routers, but given that you have few big boxes this isn't an
issue.

I am a bit weary of the idea to make 8 routers be RPs with MSDP and anycast
RP for your organizations size. This should not really be necessary unless
each of these happens to represent a geographical location that likely may
become isolated.

For reasons of simplicity and ease of troubleshooting i would recommend
to minimize the number of RP/MSDP-speaker. Purely for redundancy, two
RPs are really good enough. If you have to consider network partitioning
then you may need more, but only if the network partitions actually
have IP multicast applications that usefully can continue to run during
partition time.

Of course, there are factors invalidating this recommendation like:
- you intend to run the routers more like a lab than a production
  network, so some of them will always be down.
- you intend to run PIM-SM with SPT-threshold infinity and the
  load of traffic passing through the RPs then is an issue.
..but these typically tend to be unusal circumstances for your type
of organization ;-)


> Questions are:
> 
>   What is the prefered method these days determining RP?  Anycast-RP or
> BSR?
>  
>   Are there any problems in running Anycast-RP with HSRP?

There's no specific interaction of anycast-RP with HSRP that i am aware
of, but i've seen some crazy ways of using HSRP, which will interact badly
with PIM:

HSRP is meant to allow hosts to gain redundancy for a statically/DHCP
configured "default gateway". If you use it this way, and if none of the
RP addresses you use are HSRP addresses, then you're fine.

On the other hand, if you start using HSRP between routers, you are 
in trouble:
  - Consider two upstream routers using HSRP.
  - One (or more) downstream routers have static (default) routes to
    the HSRP address
  - PIM will fail to work in this situation, because the downstream
    routers will RPF back to the HSRP address, but they will not
    see any PIM messages from that HSRP address.

>   How hard is it to debug problems with Anycast-RP vs. PIM BSR?
> 
>   Will MSDP updates cause a big hit on our 6500 MSFCs?  

No, it will not with the amount of SA that you have in the Internet
or even twice that. Two recommendations though:

  - Try to use Cisco IOS 12.1(13)E on your routers if you want
    MSDP to scale best. We've implemented performance improvements
    there to have it scale to more SAs.
  
  - Configure "ip msdp sa-limit" on your MSDP speakers to ensure that
    the router won't suffer due to MSDP state explosion from DDoS
    attacks.

Cheers
        Toerless