All things related to multicast

["Michael H. Lambert" <>]

I'm having to jump through hoops because we don't have a real multicast 
border router for the PSC network.  The topology looks simple enough, but 
our beacons aren't showing up on the AG beacon page.

We have three relevant routers:

Router A:  A Cisco 12008 which is our GigaPoP Abilene border router in 
AS5050 and runs PIM-SM/MSDP/BGP-4+.  It runs IOS 12.0(15)S1.  It has two 
relevant interfaces:

    A1:  Point-to-point to Abilene (POS)
    A2:  Point-to-point GigE VLAN to Router B

[GigaPoP customers do the multicast suite with this router and it works.]

Router B:  A Juniper M40 which is our GigaPoP core router.  It is also in 
AS5050 and runs JunOS 5.1.  It also has two relevant interfaces:

    B1:  Point-to-point GigE VLAN to Router A (the other end of A2)
    B2:  GRE tunnel to Router C

[Our goal is to migrate all customer connections to this router.]

Router C:  A Cisco 7507 which is the PSC internal router.  Because of 
layer-eight considerations it isn't sitting behind a real multicast border 
router.  It is running IOS 12.1(5)T9.  Again, there are two relevant 
interfaces:

    C1:  GRE tunnel to Router B
    C2:  GigE VLAN (ISL trunking) to subnet with beacon servers

The PSC network is 128.182.0.0/16; the beacons are all on 128.182.61.0/24.

-----

PIM-SM is enabled on all the above interfaces.

Router C has a default mroute pointing out the GRE tunnel.  It MSDP peers 
with router B with only the standard sanity filter.  Its loopback address 
(128.182.13.13) is configured as a static RP.  Its sa-cache is empty.  A 
"show ip mroute summ" shows a number of entries for the beacon group 
(233.2.171.1).  The local entries are in state CTA which doesn't look 
unreasonable; however, the remote ones are in state CJT.  I would expect an 
"M" in there if things are working right.  I'm also showing beacon traffic 
on this router (internal and external sources).

Router B does the full suite with Router A.  It has a static mroute (in rib 
inet.2) to 128.182.0.0/16 over the tunnel.  Its SA table shows entries from 
Router C with the right peer and originator.  As far as I can tell, this 
router is doing the right thing.

Router A has a static mroute to 128.182.0.0/16 over the GigE to Router B. 
It is hearing (via MSDP) SAs for (128.182.61.*, 233.2.171.1) from Router B. 
As far as I can tell, Router A is NOT advertising our SAs to the Abilene 
core router with which it peers.  I would like it to.

Note:  The MSDP peering between Routers A and B is configured as a mesh 
group because that seems proper as they are in the same AS.

(128.182.61.*, 233.2.171.1) entries are timing out (on Router A as well as 
Abilene core routers) every three-odd minutes.  However, they seem to be 
persistent in the SA cache on these routers.

This problem looks like it should be reasonably straightforward.  My first 
guess is that I'm missing something in the MSDP configuration on A to tell 
it to redistribute SAs learned from a mesh-group peer.  Then again, maybe 
there is some profound subtlety going on which I am overlooking 
completely--as in why don't mroutes on Router A have the "M" flag set?

As a side note:  We have gotten multicast to work from our internal network 
with the Juniper out of the loop.  We've nailed up a PVC between Routers A 
and C.  This works fine for beacon traffic.  However, because of an IOS 
bug, when we put real Access Grid traffic on the PVC, we have a choice of 
heavy packet loss on the ATM interface (with mroute-cache enabled on the 
GigE) or 99% CPU load (with mroute-cache disabled).  Thus, we would like to 
get this to work through the Juniper.

Any thoughts are most welcome.

Thanks,

Michael

+-----------------------------------------------------------------------+
| Michael H. Lambert, Network Engineer           Phone: +1 412 268-4960 |
| Pittsburgh Supercomputing Center               FAX:   +1 412 268-8200 |
| 4400 Fifth Avenue, Pittsburgh, PA  15213       

        |
+-----------------------------------------------------------------------+