[c-nsp] Multicast distribution over backbone
Phil Mayers
p.mayers at imperial.ac.uk
Wed Dec 1 09:31:41 EST 2010
On 01/12/10 13:31, Alexander Clouter wrote:
>> Designate one router as the RP. Configure PIM-SM each router. Configure
>> them all with the same RP.
>>
>> You only need MSDP to pass source info between PIM-SM RPs.
>>
> All our core routers are RPs with the same 'anycasted' address so we
> have resilence. We use MSDP to then make sure everything is then
> sync'ed up:
Sure. We do the same thing with our RPs (although we make use of the
JunOS anycast-rp stuff to avoid having to run MSDP). Best to get a
single RP working first though ;o)
One thing: Personally I'd be a bit cautious about making *every* router
an RP. One unfortunate side effect of being an RP is the CPU-processing
of the encaps/decaps traffic. This is particularly noticeable if you
don't have:
ip pim [vrf NAME] register-rate-limit N
...on every *other* router, as high-bandwidth transmitter can punt lots
of packets/sec to the RP CPU for brief periods until the (s,g) state is
built.
For this reason we use separate routers as RPs. Our main RPs are two
Juniper M7i (which have tunnel services PICs so to the encaps/decaps in
hardware) but we also use an old Cisco 2691 with appropriate IOS as the
RP for 239.0.0.0/8 inside our firewall, for boring reasons relating to
lack of jumbo frames on the firewalls, fragmentation of PIM register
packets containing full-size packets, and differing IOS/JunOS behaviour.
Obviously it depends on your traffic patterns and which routers you're
using. There is no one-size-fits-all.
More information about the cisco-nsp
mailing list