Auto-RP with only Sparse Mode

From: Bob Sinclair (bsinclair@netmasterclass.net)
Date: Thu Jun 17 2004 - 18:19:54 GMT-3


All,

A few folks have lately mentioned they have been able to distribute Auto-RP
mappings with only Sparse Mode configured on the interfaces. I have
duplicated that result without using the ip pim autorp listener command.
Would appreciate any comments. Topology is as follows. IOS versions are
12.2(4)T7 and 12.2(15)T11.

R5(source)
-----------R3-----------R4(rp-candidate)------------R1(MA)--------R2---------
-R6

R3, R4, R1, R2 and R6 each have sparse-mode configured on their external
interfaces and on a loopback. The loopback also is joined to group 231.1.1.1.
R4 is configured as an RP-candidate to its loopback 172.16.104.1, and R1 is
configured as a mapping agent using loopback 172.16.101.1.

What I found was that R4, R1, R2 and R6 all learned an RP-mapping. R3 did
not. It was most surprising to me that R6 would learn it, since R2 would have
to somehow forward the information, and it was not supposed to be distributing
224.0.1.40 in dense mode.

On R6, I compared the output of debug ip pim auto-rp with the result of a
packet capture on the R2-R6 Ethernet link. What I found was that R2 forwarded
to R6 a PIM V1 RP-Reachable message! This message had an IP source address of
172.16.104.1, and an IP destination address of 224.0.0.2, the all routers
address. Surprisingly, both R1 and R2 forwarded this message, even though
the range 224.0.0.X is supposed to be link-local. Beau Williamson mentions
the PIM V1 RP-Reachable message on page 540 of his book Developing IP
Multicast Networks. In PIM V1, the RP sends these messages down the shared
tree as a kind of RP keepalive. Apparently, PIM routers forward these, though
there is no mroute entry for the group.

Note that R3 did not receive any of these messages from the RP, and never
learned the RP automatically. This is apparently explained by the fact that
R3 is on the SPT between the RP and the source, not on the shared tree. In
order to get an end-end ping from R5 to R6, I had to do a static RP-address
command on R3 pointing to R4. Then the pings worked. Also, once R3 had a
static RP entry, it was able to additionally learn the Auto-RP mapping -
224.0.1.40 showed up as a sparse-mode group!! Once R6 learned the RP from the
PIM V1 RP-Reachable message, it then obtained an Auto-RP message, and
224.0.1.40 shows up as a Dense mode group! There was quite a bit of latency
to the process - the RP-Reachable messages seem to have an infrequent period -
and I would not rely on it in production.

All of which is to say that "a fact is the conclusion you hold when you stop
investigating," (a paraphrase of Richard BenVeniste). Does Auto-RP require
sparse-dense? Well, how deep do you want to go into that question? If a lab
requires something like "a dynamic RP announcement method that requires
sparse-dense", I would go for Auto-RP!

Bob Sinclair
CCIE #10427, CISSP, MCSE
www.netmasterclass.net



This archive was generated by hypermail 2.1.4 : Sat Jul 03 2004 - 19:40:43 GMT-3